Reflection AI
reflection.ai
Automated, point-in-time. Not independently reviewed or approved by TRUSTYCYBER unless expressly stated.
- No clear commitment that your data will not train their models
- Data retention window not stated
See Before you sign, with what to ask for ↓
Scanned Oct 21, 2026 · Public evidence · Point-in-time
Higher scores indicate stronger, independently supported AI assurance, governance, AI systems, models, data, supply chain and third-party assurance evidence. The score and the decision are related but different: a well-scored vendor can still carry conditions.
Before you sign
Why it matters: The public sources scanned do not clearly state that customer prompts, files, and outputs are excluded from training or fine-tuning of first- or third-party models.
What to ask for: Get a no-training clause covering first- and third-party models into the DPA/order form.
“Inputs and Outputs. If you interact with Reflection-hosted instances to our models, such as through a Reflection-managed API, we collect the content you submit (“Inputs”) and the responses generated (“Outputs”). If you include personal data in your Inputs, we will collect it and it may be reproduced in your Outputs. Depending on your choices, we may use Inputs and Outputs to train and improve our models.”
“Unless required by applicable law to which Reflection is subject, Reflection will only process Customer Personal Data to provide or maintain the Services, and in compliance with Customer’s documented instructions (including as set out in the Agreement and this DPA).”
Why it matters: The public sources scanned do not give a retention period for customer data or prompts/outputs.
What to ask for: Get retention windows, in days, in writing.
“Within thirty (30) days of the date of termination or expiration of the Agreement, Reflection will, at the choice of Customer, delete or return to Customer all copies of Customer Data (including Customer Personal Data) processed by Reflection or any Subprocessors, except to the extent (i) Applicable Data Protection Laws or other applicable legal or regulatory requirements requires storage of the Customer Data, (ii) retention of the Customer Data by Reflection is necessary to resolve a dispute between the parties, or (iii) retention of the Customer Data is necessary to combat harmful use of the Services.”
“We retain personal data only for as long as we need it in order to operate our Services, or for other legitimate business purposes, such as resolving disputes, safety and security reasons, or complying with our legal obligations. How long we retain personal data depends on the type of data and how we use it.”
Buyer questionsanswered only from the public evidence scanned, click a row for the answer
!Will they train on your data?Ask the vendor
For Reflection-hosted model instances such as the Reflection-managed API, Reflection collects Inputs and Outputs and states that, depending on the user's choices, it may use them to train and improve its models. Training use therefore exists and is choice-dependent; the default setting and the opt-out mechanism are not described.
Requires written confirmation, see Before you sign ↓
“Inputs and Outputs. If you interact with Reflection-hosted instances to our models, such as through a Reflection-managed API, we collect the content you submit (“Inputs”) and the responses generated (“Outputs”). If you include personal data in your Inputs, we will collect it and it may be reproduced in your Outputs. Depending on your choices, we may use Inputs and Outputs to train and improve our models.”
“Unless required by applicable law to which Reflection is subject, Reflection will only process Customer Personal Data to provide or maintain the Services, and in compliance with Customer’s documented instructions (including as set out in the Agreement and this DPA).”
!How long do they keep your data?Ask the vendor
Within 30 days of termination or expiry, Reflection will delete or return all copies of Customer Data held by it or its subprocessors, at the customer's choice, with carve-outs for legal retention requirements, dispute resolution and combating harmful use of the Services.
Requires written confirmation, see Before you sign ↓
“Within thirty (30) days of the date of termination or expiration of the Agreement, Reflection will, at the choice of Customer, delete or return to Customer all copies of Customer Data (including Customer Personal Data) processed by Reflection or any Subprocessors, except to the extent (i) Applicable Data Protection Laws or other applicable legal or regulatory requirements requires storage of the Customer Data, (ii) retention of the Customer Data by Reflection is necessary to resolve a dispute between the parties, or (iii) retention of the Customer Data is necessary to combat harmful use of the Services.”
“We retain personal data only for as long as we need it in order to operate our Services, or for other legitimate business purposes, such as resolving disputes, safety and security reasons, or complying with our legal obligations. How long we retain personal data depends on the type of data and how we use it.”
!Who else can access your data?Ask the vendor
Reflection publishes a named subprocessor list for its API with location and purpose. Inference is performed by Together AI (Canada) and Voltage Park, Inc. (United States) with Tailscale providing inference routing; Google Cloud Platform supplies cloud infrastructure; Cloudflare is the CDN; WorkOS, Intercom, Stripe, Dow Jones, DataGrail, Cinder Technologies and Bird B.V. cover identity, support, payments, sanctions screening, privacy requests, content moderation and customer service. Bird processes in the US, Belgium, Ireland, India, Netherlands and UK.
Confirm in writing: Ask the vendor to state this in writing before signing.
“Google Cloud Platform API United States Cloud infrastructure Intercom API United States Customer service Stripe API United States Payment Processing Dow Jones API United States Export and sanctions compliance Tailscale API United States Inference routing Together AI API Canada Inference DataGrail API Website United States Data Subject Rights Request Cinder Technologies, Inc. API United States Content moderation Bird B.V. API United States, Belgium, Ireland, India, Netherlands, and UK Customer service Voltage Park, Inc. API United States Inference”
!Where is your data processed?Ask the vendor
The DPA incorporates EU SCCs Module Two and Module Three by reference, with a UK Addendum and a Swiss Addendum, Irish governing law and forum for the SCCs, and general written authorisation for subprocessors. Customer data is processed in the United States and at subprocessor locations; there is no in-region residency option.
Confirm in writing: Ask the vendor to state this in writing before signing.
“The parties agree that, to the extent required by Applicable Data Protection Laws, the terms of the SCCs Module Two (controller to processor) and/or Module Three (processor to processor), as completed as described in Schedule 3 of this DPA, are hereby incorporated by reference and will be deemed to have been executed by the parties.”
“Reflection AI processes personal data for the purposes described in this policy on servers located in various jurisdictions, including processing and storing your personal data in our facilities and servers in the United States, or in countries or territories where our affiliates and partners or our vendors and service providers are located. While data protection law varies by country, we apply the protections described in this policy to your personal data regardless of where it is processed, and only transfer that data pursuant to legally valid transfer mechanisms.”
✓What happens in a security incident?Clear
Reflection commits to notify the customer in writing without undue delay and within 72 hours of becoming aware of a Security Breach, and to cooperate with the customer's investigation, providing the nature of the breach, affected categories and record counts, likely consequences and mitigation measures.
“Reflection will notify Customer in writing without undue delay, but in any event within 72 hours, after becoming aware of any Security Breach, and will assist Customer in complying with Customer’s obligations under Applicable Data Protection Laws by reasonably cooperating with Customer’s investigation of the Security Breach.”
“Reflection’s incident management procedures are designed to allow Reflection to investigate, respond to, mitigate, and notify of events related to Reflection’s technology and information assets that process Customer Data. These incident response procedures are tested annually.”
Email to send the vendor7 items to confirm in writing
Hello Reflection AI team, We are assessing Reflection AI as part of our supplier review. Before we proceed, please confirm the following in writing: 1. Do you use our data (inputs, outputs or uploaded files) to train or improve your models? Please confirm in writing, including any opt-out and whether it applies to every plan. 2. How long do you retain our data, in days, for each type (inputs, outputs, logs and backups), and how is it deleted? 3. Please provide your current, dated subprocessor list and explain how you notify customers of changes. 4. In which countries and regions is our data processed and stored, and can processing be limited to a region we choose? 5. Which certifications or audit reports (SOC 2 Type II, ISO 27001, ISO 42001) does Reflection hold for the hosted API, and who performed the external third-party assurance evaluations referenced in the Beam model card? 6. Stripe appears to be involved as a service provider: confirm whether this dependency exists, and whether it processes customer data. 7. Microsoft appears to be involved as a service provider: confirm whether this dependency exists, and whether it processes customer data. A written reply to each point, or a link to where it is documented, is enough. Where a point is covered by a certification or independent report, please include the certificate number or the report and its date. Thank you,
Key findingsclick a row for the evidence
✓Enterprise DPA with named subprocessors, 72-hour breach notice and 30-day deletionStrong
Reflection publishes a commercial DPA that incorporates EU SCCs (Modules Two and Three) with UK and Swiss addenda, commits to breach notification within 72 hours, deletion or return of Customer Data within 30 days of termination, AES-256 at rest and TLS in transit, and a 15-day subprocessor objection right. A separate subprocessor register names every provider with location and purpose.
For a vendor whose first model launched this month, this is a mature contractual baseline. A buyer using the hosted API can rely on contractual terms rather than privacy-policy language, and can map every third party in the request path before signing.
“Within thirty (30) days of the date of termination or expiration of the Agreement, Reflection will, at the choice of Customer, delete or return to Customer all copies of Customer Data (including Customer Personal Data) processed by Reflection or any Subprocessors, except to the extent (i) Applicable Data Protection Laws or other applicable legal or regulatory requirements requires storage of the Customer Data, (ii) retention of the Customer Data by Reflection is necessary to resolve a dispute between the parties, or (iii) retention of the Customer Data is necessary to combat harmful use of the Services.”
“Google Cloud Platform API United States Cloud infrastructure Intercom API United States Customer service Stripe API United States Payment Processing Dow Jones API United States Export and sanctions compliance Tailscale API United States Inference routing Together AI API Canada Inference DataGrail API Website United States Data Subject Rights Request Cinder Technologies, Inc. API United States Content moderation Bird B.V. API United States, Belgium, Ireland, India, Netherlands, and UK Customer service Voltage Park, Inc. API United States Inference”
“In the event that Reflection wishes to appoint an additional Subprocessor: (a) Reflection will provide Customer reasonable notice of the new Subprocessor prior to giving the Subprocessor access to Customer Personal Data; and (b) Customer may, on the basis of reasonable data privacy or data security concerns, object to Reflection’s use of such Subprocessor by providing Reflection with written notice of the objection within fifteen (15) days of the date of such notice, or Customer is deemed to consent to the new Subprocessor.”
“Reflection will notify Customer in writing without undue delay, but in any event within 72 hours, after becoming aware of any Security Breach, and will assist Customer in complying with Customer’s obligations under Applicable Data Protection Laws by reasonably cooperating with Customer’s investigation of the Security Breach.”
“Reflection utilizes industry standard encryption methods for protection of Customer Data, including a minimum of AES-256 for data at rest, and TLS for Customer Data in transit over public networks.”
“The parties agree that, to the extent required by Applicable Data Protection Laws, the terms of the SCCs Module Two (controller to processor) and/or Module Three (processor to processor), as completed as described in Schedule 3 of this DPA, are hereby incorporated by reference and will be deemed to have been executed by the parties.”
!Hosted API inputs and outputs may be used for training, with the default and opt-out undisclosedGap
The Privacy Policy states that for Reflection-hosted instances, such as the Reflection-managed API, inputs and outputs may be used to train and improve models depending on the user's choices, but does not say what the default is or how the choice is made. The enterprise DPA limits processing to providing and maintaining the Services under documented instructions and excludes use outside the business relationship, which implies training is not a permitted purpose, but it never says so in terms. Self-hosted Beam is outside this question entirely since Reflection never sees the data.
Whether prompts and completions feed future Reflection models is the single most consequential data question for an API buyer. Right now the answer differs by product surface and rests on an implied reading of the DPA rather than an explicit no-training clause.
Question for vendor: For API keys issued under the Commercial Terms, are Inputs and Outputs ever used to train or improve Reflection models? What is the default setting for a new account, where is the opt-out, and will Reflection add an explicit no-training commitment to the DPA?
“Inputs and Outputs. If you interact with Reflection-hosted instances to our models, such as through a Reflection-managed API, we collect the content you submit (“Inputs”) and the responses generated (“Outputs”). If you include personal data in your Inputs, we will collect it and it may be reproduced in your Outputs. Depending on your choices, we may use Inputs and Outputs to train and improve our models.”
“This Privacy Policy does not apply to content that we process on behalf of customers of our enterprise business offerings. Our use of that data is governed by our customer agreements covering access to and use of those offerings.”
“Unless required by applicable law to which Reflection is subject, Reflection will only process Customer Personal Data to provide or maintain the Services, and in compliance with Customer’s documented instructions (including as set out in the Agreement and this DPA).”
!Inference for the hosted API is performed by third-party providers in Canada and the USGap
The subprocessor register lists Together AI (Canada) and Voltage Park, Inc. (United States) for inference and Tailscale for inference routing, with Google Cloud Platform as cloud infrastructure. Customer prompts sent to the Reflection API are therefore processed on infrastructure operated by at least two external inference providers, and the routing rule between them is not described. This applies only to the hosted API; the subprocessors_and_supply_chain and customer_data_treatment domains are assessed as covered and partial rather than not_applicable because that hosted chain exists beside the open weights.
A buyer who assumed Reflection ran its own inference would be wrong. Each inference subprocessor is a point where prompt content could be logged or retained, and a Canadian processing location matters for buyers with residency constraints.
Question for vendor: Which requests route to Together AI versus Voltage Park, can a customer pin inference to one provider or region, and what are those providers' prompt logging and retention settings for Reflection traffic?
“Google Cloud Platform API United States Cloud infrastructure Intercom API United States Customer service Stripe API United States Payment Processing Dow Jones API United States Export and sanctions compliance Tailscale API United States Inference routing Together AI API Canada Inference DataGrail API Website United States Data Subject Rights Request Cinder Technologies, Inc. API United States Content moderation Bird B.V. API United States, Belgium, Ireland, India, Netherlands, and UK Customer service Voltage Park, Inc. API United States Inference”
“The parties agree that, to the extent required by Applicable Data Protection Laws, the terms of the SCCs Module Two (controller to processor) and/or Module Three (processor to processor), as completed as described in Schedule 3 of this DPA, are hereby incorporated by reference and will be deemed to have been executed by the parties.”
“Reflection AI processes personal data for the purposes described in this policy on servers located in various jurisdictions, including processing and storing your personal data in our facilities and servers in the United States, or in countries or territories where our affiliates and partners or our vendors and service providers are located. While data protection law varies by country, we apply the protections described in this policy to your personal data regardless of where it is processed, and only transfer that data pursuant to legally valid transfer mechanisms.”
!No named independent certification or audit reportGap
No SOC 2, ISO 27001, ISO 42001 or equivalent certification is named in any collected document. The DPA promises certifications or reports on request without identifying any, Schedule 2 describes only internal audit and risk procedures reporting to senior management, and the model card refers to unnamed external third-party assurance evaluations of Beam's safety. The independent_assurance domain is marked partial rather than not_evidenced because a contractual audit right and an external safety evaluation are both disclosed; it is not not_applicable, since an audit of either the hosted service or the model release process is clearly possible.
Without a named attestation the security controls in Schedule 2 are self-described. A buyer cannot verify the encryption, access control or incident testing claims, and the safety evaluation results cannot be attributed to an identifiable independent party.
Question for vendor: Which certifications or audit reports (SOC 2 Type II, ISO 27001, ISO 42001) does Reflection hold for the hosted API, and who performed the external third-party assurance evaluations referenced in the Beam model card?
“Before releasing Beam, we ran a comprehensive set of safety evaluations to measure its performance and identify potential risks requiring mitigation. We run both Open Safety Framework evaluations, external third party assurance evaluations, and The Iris Safety Evaluations (ISE), which is Reflection’s in-house soon-to-be open-sourced safety evaluation suite, combining single-turn, multi-turn, jailbreak, and agentic evaluations to measure adherence to our safety policies.”
“Upon Customer’s written request at reasonable intervals, and subject to reasonable confidentiality controls, Reflection shall, either (i) make available for Customer’s review copies of certifications or reports demonstrating Reflection’s compliance with prevailing data security standards applicable to the processing of Customer Personal Data, or (ii) if the provision of reports or certifications pursuant to (i) is not reasonably sufficient under Privacy Laws, allow Customer’s independent third party representative to conduct an audit or inspection of Reflection’s data security infrastructure and procedures”
“Reflection defines audit and risk assessment procedures for the purposes of: (a) periodic review and assessment of risks to Reflection’s organization, assets, and data, (b) monitoring and maintaining compliance with Reflection’s policies and procedures, and (c) reporting the condition of its information security and compliance to internal senior management.”
✓Transparent model documentation and safety evaluation reporting for BeamStrong
The Beam model card discloses architecture, training data provenance, knowledge cutoff, benchmark methodology and quantitative safety evaluation results across all released numerics, and the launch post describes an iterative adversarial red-teaming and safety RL process. The licence explicitly permits adversarial evaluation and security research, and Reflection's crawler honours robots.txt with published IP ranges.
A buyer self-hosting Beam has enough published detail to run its own evaluation and to understand what the model was trained on. The permissive stance on adversarial testing lowers the barrier to independent scrutiny.
“Beam is an open model built by Reflection. Beam takes text as input and generates text as output. This release includes the instruction-tuned model in BF16, FP8, and NVFP4 numerics. Beam uses a sparse mixture-of-experts (MoE) architecture with 501B total parameters and 23B active parameters. It supports a context window of 1M tokens and was developed predominantly for English.”
“Beam is pretrained on a diverse variety of textual data from publicly available sources and data obtained through commercial partner agreements. The corpus includes general web content, source code, technical documentation, reference materials, and educational and scientific documents.”
“Iris Safety Evaluations (Combined All Harm Areas) Table 1 Row label Beam (BF16) Beam (FP8) Beam (NVFP4) Harmful Requests: Refusal/Pivot Rate 88.9% 89.7% 88.5% Harmless Requests: Compliance Rate 81.5% 81.6% 82.1%”
“The dataset was built adversarially and iteratively: in each round, we trained a model, generated prompts that elicited harmful or over-refusing behavior from it, and folded the successful attacks back into SFT mixture for the next round. For safety RL, we similarly sourced single-turn, multi-turn, jailbreak, and agentic scenarios in which a simulated adversary pressures a tool-using model to take unsafe actions, to simultaneously reduce over-refusals and harmful compliance.”
“Reflection does not prohibit distillation, use of outputs to train other models, fine-tuning, parameter-efficient adaptation, quantization, merging, defensive security research on systems you are authorized to test, and good-faith adversarial evaluation of the model.”
“If you would like to limit or stop Reflectionbot's crawling, the best way to do that is through robots.txt rather than blocking requests with errors like HTTP 403 . Reflectionbot reads and honors robots.txt , and changes are typically picked up within about a day.”
?Technical dependency observed: StripeObservation
Technical evidence (DNS, certificate, or HTTP) shows a dependency on Stripe as a service provider. This dependency is not identified in the vendor's published materials.
An undisclosed provider dependency is exactly the gap independent verification exists to surface.
Question for vendor: Verification required — Stripe appears to be involved as a service provider: confirm whether this dependency exists, and whether it processes customer data.
“Beam is pretrained on a diverse variety of textual data from publicly available sources and data obtained through commercial partner agreements. The corpus includes general web content, source code, technical documentation, reference materials, and educational and scientific documents.”
“This month, we will release the weights under an Apache 2.0 license, along with documentation and the full stack for running, evaluating, and fine-tuning the model.”
“Google Cloud Platform API United States Cloud infrastructure Intercom API United States Customer service Stripe API United States Payment Processing Dow Jones API United States Export and sanctions compliance Tailscale API United States Inference routing Together AI API Canada Inference DataGrail API Website United States Data Subject Rights Request Cinder Technologies, Inc. API United States Content moderation Bird B.V. API United States, Belgium, Ireland, India, Netherlands, and UK Customer service Voltage Park, Inc. API United States Inference”
“In the event that Reflection wishes to appoint an additional Subprocessor: (a) Reflection will provide Customer reasonable notice of the new Subprocessor prior to giving the Subprocessor access to Customer Personal Data; and (b) Customer may, on the basis of reasonable data privacy or data security concerns, object to Reflection’s use of such Subprocessor by providing Reflection with written notice of the objection within fifteen (15) days of the date of such notice, or Customer is deemed to consent to the new Subprocessor.”
“In addition, because we release our models publicly as open-weight models, those models are deployed and operated by third parties, and we do not control the outputs they generate.”
?Technical dependency observed: MicrosoftObservation
Technical evidence (DNS, certificate, or HTTP) shows a dependency on Microsoft as a service provider. This dependency is not identified in the vendor's published materials.
An undisclosed provider dependency is exactly the gap independent verification exists to surface.
Question for vendor: Verification required — Microsoft appears to be involved as a service provider: confirm whether this dependency exists, and whether it processes customer data.
“Beam is pretrained on a diverse variety of textual data from publicly available sources and data obtained through commercial partner agreements. The corpus includes general web content, source code, technical documentation, reference materials, and educational and scientific documents.”
“This month, we will release the weights under an Apache 2.0 license, along with documentation and the full stack for running, evaluating, and fine-tuning the model.”
“Google Cloud Platform API United States Cloud infrastructure Intercom API United States Customer service Stripe API United States Payment Processing Dow Jones API United States Export and sanctions compliance Tailscale API United States Inference routing Together AI API Canada Inference DataGrail API Website United States Data Subject Rights Request Cinder Technologies, Inc. API United States Content moderation Bird B.V. API United States, Belgium, Ireland, India, Netherlands, and UK Customer service Voltage Park, Inc. API United States Inference”
“In the event that Reflection wishes to appoint an additional Subprocessor: (a) Reflection will provide Customer reasonable notice of the new Subprocessor prior to giving the Subprocessor access to Customer Personal Data; and (b) Customer may, on the basis of reasonable data privacy or data security concerns, object to Reflection’s use of such Subprocessor by providing Reflection with written notice of the objection within fifteen (15) days of the date of such notice, or Customer is deemed to consent to the new Subprocessor.”
“In addition, because we release our models publicly as open-weight models, those models are deployed and operated by third parties, and we do not control the outputs they generate.”
Assurance dimensionsweighted components of one score, not eight separate ratings
Organisation & AI governance15% of the score40
Who owns AI risk: policies, responsible-AI principles, human oversight and administrative controls.
Assessed against ISO 42001 · NIST AI RMF · AIUC-1
“Reflection utilizes industry standard encryption methods for protection of Customer Data, including a minimum of AES-256 for data at rest, and TLS for Customer Data in transit over public networks.”
“If you would like to limit or stop Reflectionbot's crawling, the best way to do that is through robots.txt rather than blocking requests with errors like HTTP 403 . Reflectionbot reads and honors robots.txt , and changes are typically picked up within about a day.”
“To develop and train the artificial intelligence models including those that we release publicly as open-weight models, including by processing information that is publicly available on the internet (which may include personal data). Our legitimate interests (Article 6(1)(f) GDPR) in developing, improving, and releasing safe and capable artificial intelligence models. We apply safeguards when processing this information, and you may object to this processing at any time (see “Your Rights” below).”
“Reflection defines audit and risk assessment procedures for the purposes of: (a) periodic review and assessment of risks to Reflection’s organization, assets, and data, (b) monitoring and maintaining compliance with Reflection’s policies and procedures, and (c) reporting the condition of its information security and compliance to internal senior management.”
AI system10% of the score53
What AI the vendor actually runs and where it sits in the product, the map everything else is judged against.
Assessed against ISO 42001 · NIST AI RMF
“Beam is an open model built by Reflection. Beam takes text as input and generates text as output. This release includes the instruction-tuned model in BF16, FP8, and NVFP4 numerics. Beam uses a sparse mixture-of-experts (MoE) architecture with 501B total parameters and 23B active parameters. It supports a context window of 1M tokens and was developed predominantly for English.”
“API Call Beam over HTTPS with the Reflection API or the official OpenAI SDKs. Mirror Code in your terminal with Mirror, Reflection’s coding agent built on Beam, or set up another coding agent to use Beam. Deployment Run Beam on your own hardware and verify the result.”
“In addition, because we release our models publicly as open-weight models, those models are deployed and operated by third parties, and we do not control the outputs they generate.”
How AI features are evaluated before and after release, evals, red-teaming, monitoring.
Assessed against NIST GenAI · NIST AI RMF · ISO 42001
“Iris Safety Evaluations (Combined All Harm Areas) Table 1 Row label Beam (BF16) Beam (FP8) Beam (NVFP4) Harmful Requests: Refusal/Pivot Rate 88.9% 89.7% 88.5% Harmless Requests: Compliance Rate 81.5% 81.6% 82.1%”
“The dataset was built adversarially and iteratively: in each round, we trained a model, generated prompts that elicited harmful or over-refusing behavior from it, and folded the successful attacks back into SFT mixture for the next round. For safety RL, we similarly sourced single-turn, multi-turn, jailbreak, and agentic scenarios in which a simulated adversary pressures a tool-using model to take unsafe actions, to simultaneously reduce over-refusals and harmful compliance.”
How changes to models and AI features are controlled and communicated: versioning, notice, rollback.
Assessed against ISO 27001 · ISO 42001
“Reflection’s change management procedures and tracking mechanisms are designed to test, approve, and monitor all changes to Reflection’s technology and information assets that process Customer Data.”
“Reflection may update these measures from time to time, provided that such updates do not materially reduce the overall security of the Services.”
AI system description: vendor-evidenced, not yet independently corroborated.
Testing & evaluation: vendor-evidenced, not yet independently corroborated.
Model10% of the score60
Which model providers and versions process customer data, and how changes to them are disclosed.
Assessed against NIST GenAI · ISO 42001
“Beam is pretrained on a diverse variety of textual data from publicly available sources and data obtained through commercial partner agreements. The corpus includes general web content, source code, technical documentation, reference materials, and educational and scientific documents.”
“This month, we will release the weights under an Apache 2.0 license, along with documentation and the full stack for running, evaluating, and fine-tuning the model.”
“In addition, because we release our models publicly as open-weight models, those models are deployed and operated by third parties, and we do not control the outputs they generate.”
Model provider transparency: vendor-evidenced, not yet independently corroborated.
Agent10% of the score0
No independently corroborated agent-specific assurance (AIUC-class or equivalent).
Customer data15% of the score40
Whether customer data trains models, how long it is retained, where it is processed, and how it is protected.
Assessed against ISO 27018 · OAIC APPs · EU AI Act
“Inputs and Outputs. If you interact with Reflection-hosted instances to our models, such as through a Reflection-managed API, we collect the content you submit (“Inputs”) and the responses generated (“Outputs”). If you include personal data in your Inputs, we will collect it and it may be reproduced in your Outputs. Depending on your choices, we may use Inputs and Outputs to train and improve our models.”
“This Privacy Policy does not apply to content that we process on behalf of customers of our enterprise business offerings. Our use of that data is governed by our customer agreements covering access to and use of those offerings.”
“Unless required by applicable law to which Reflection is subject, Reflection will only process Customer Personal Data to provide or maintain the Services, and in compliance with Customer’s documented instructions (including as set out in the Agreement and this DPA).”
“Within thirty (30) days of the date of termination or expiration of the Agreement, Reflection will, at the choice of Customer, delete or return to Customer all copies of Customer Data (including Customer Personal Data) processed by Reflection or any Subprocessors, except to the extent (i) Applicable Data Protection Laws or other applicable legal or regulatory requirements requires storage of the Customer Data, (ii) retention of the Customer Data by Reflection is necessary to resolve a dispute between the parties, or (iii) retention of the Customer Data is necessary to combat harmful use of the Services.”
“We retain personal data only for as long as we need it in order to operate our Services, or for other legitimate business purposes, such as resolving disputes, safety and security reasons, or complying with our legal obligations. How long we retain personal data depends on the type of data and how we use it.”
“Reflection AI processes personal data for the purposes described in this policy on servers located in various jurisdictions, including processing and storing your personal data in our facilities and servers in the United States, or in countries or territories where our affiliates and partners or our vendors and service providers are located. While data protection law varies by country, we apply the protections described in this policy to your personal data regardless of where it is processed, and only transfer that data pursuant to legally valid transfer mechanisms.”
AI supply chain10% of the score60
The named chain of AI and infrastructure providers behind the product, and how it is disclosed.
Assessed against ISO 27001 · ISO 27017
“Google Cloud Platform API United States Cloud infrastructure Intercom API United States Customer service Stripe API United States Payment Processing Dow Jones API United States Export and sanctions compliance Tailscale API United States Inference routing Together AI API Canada Inference DataGrail API Website United States Data Subject Rights Request Cinder Technologies, Inc. API United States Content moderation Bird B.V. API United States, Belgium, Ireland, India, Netherlands, and UK Customer service Voltage Park, Inc. API United States Inference”
“In the event that Reflection wishes to appoint an additional Subprocessor: (a) Reflection will provide Customer reasonable notice of the new Subprocessor prior to giving the Subprocessor access to Customer Personal Data; and (b) Customer may, on the basis of reasonable data privacy or data security concerns, object to Reflection’s use of such Subprocessor by providing Reflection with written notice of the objection within fifteen (15) days of the date of such notice, or Customer is deemed to consent to the new Subprocessor.”
Subprocessors & supply chain: vendor-evidenced, not yet independently corroborated.
Security foundation10% of the score60
How security problems are found, reported and handled: disclosure channels, bug bounty, incident response.
Assessed against ISO 27001 · NIST AI RMF
“Reflection will notify Customer in writing without undue delay, but in any event within 72 hours, after becoming aware of any Security Breach, and will assist Customer in complying with Customer’s obligations under Applicable Data Protection Laws by reasonably cooperating with Customer’s investigation of the Security Breach.”
“Report suspected misuse, safety concerns or security vulnerabilities to [email protected] .”
“Reflection’s incident management procedures are designed to allow Reflection to investigate, respond to, mitigate, and notify of events related to Reflection’s technology and information assets that process Customer Data. These incident response procedures are tested annually.”
Vulnerability & incident handling: vendor-evidenced, not yet independently corroborated.
Independent assurance evidence10% of the score15
Whether claimed certifications and attestations are independently corroborated, checked against official registries and issuing bodies. Which certifications, and what each one does and does not establish, is stated per certificate.
Assessed against ISO 17021-1 · ISO 42006 · ISO 27006-1 · ISO 27008
“Before releasing Beam, we ran a comprehensive set of safety evaluations to measure its performance and identify potential risks requiring mitigation. We run both Open Safety Framework evaluations, external third party assurance evaluations, and The Iris Safety Evaluations (ISE), which is Reflection’s in-house soon-to-be open-sourced safety evaluation suite, combining single-turn, multi-turn, jailbreak, and agentic evaluations to measure adherence to our safety policies.”
“Upon Customer’s written request at reasonable intervals, and subject to reasonable confidentiality controls, Reflection shall, either (i) make available for Customer’s review copies of certifications or reports demonstrating Reflection’s compliance with prevailing data security standards applicable to the processing of Customer Personal Data, or (ii) if the provision of reports or certifications pursuant to (i) is not reasonably sufficient under Privacy Laws, allow Customer’s independent third party representative to conduct an audit or inspection of Reflection’s data security infrastructure and procedures”
None of the corroborated certifications is AI-specific (ISO/IEC 42001 or AIUC-class) — this assurance is security and data-transfer attestation, not AI-management-system assurance.
Legal & contractual10% of the score60
The contractual backbone a buyer can rely on, terms, DPAs, privacy commitments and regulatory posture.
Assessed against EU AI Act · OAIC APPs
“The parties agree that, to the extent required by Applicable Data Protection Laws, the terms of the SCCs Module Two (controller to processor) and/or Module Three (processor to processor), as completed as described in Schedule 3 of this DPA, are hereby incorporated by reference and will be deemed to have been executed by the parties.”
“Reflection does not prohibit distillation, use of outputs to train other models, fine-tuning, parameter-efficient adaptation, quantization, merging, defensive security research on systems you are authorized to test, and good-faith adversarial evaluation of the model.”
“In the event that Reflection wishes to appoint an additional Subprocessor: (a) Reflection will provide Customer reasonable notice of the new Subprocessor prior to giving the Subprocessor access to Customer Personal Data; and (b) Customer may, on the basis of reasonable data privacy or data security concerns, object to Reflection’s use of such Subprocessor by providing Reflection with written notice of the objection within fifteen (15) days of the date of such notice, or Customer is deemed to consent to the new Subprocessor.”
Legal & contractual transparency: vendor-evidenced, not yet independently corroborated.
What would strengthen assuranceeach figure is from today's score, resolving one changes the others
Resolving an evidence gap strengthens the dimension it belongs to, and with it the score: 48 → up to 70 with every identified gap resolved.
AI supply chainclick a node to focus it · drag to pan · zoom with the controls
View as list
What you inheritTrustyCyber's AI Shared Responsibility Matrix v1.0
This scan names the vendor’s AI supply chain without assessing the responsibility split, that depends on how you consume the product. Under the common managed models, a deployer inherits 16 of the matrix’s 41 responsibilities from providers, each with named evidence owed.
Map your inherited responsibilitiesAssurance evidence: certifications 1, registry checks and verification ladders, click to view
Not claimed in any public source scanned. This is the AI-management-system certification — in its absence, the vendor’s AI governance rests on its general security and privacy certifications.
Sources 12, click to view
Appendix: standards landscape 8, click to view
Where each standard behind this assessment stands today, from TrustyCyber’s continuously maintained standards radar.
Want to go further?
This scan assesses Reflection AI at an organisational level. Assurance can vary significantly between products, models and agents, enter the product you’re evaluating to run a deeper AI System Assurance Scan.
Scan a specific product →This automated result is based on publicly available information at the time of scanning. It is not an audit, certification, legal opinion or assurance engagement. Publicly unavailable evidence may materially change the conclusion.
Monitor for changes
Get an email if its TRUSTYCYBER Score, grade or certifications change. One credit per refresh; reading the report always stays free.