Llama
meta.com
Automated, point-in-time. Not independently reviewed or approved by TRUSTYCYBER unless expressly stated.
- No warranty, no liability, and no third-party assurance over the model
- No independent assurance evidenced
See Before you sign, with what to ask for ↓
Scanned Sep 3, 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 Llama Materials and their outputs are supplied as-is with every warranty disclaimed and liability excluded, and the licensee is made solely responsible for judging their fitness. Nothing in the collected public surface offers any independent assurance to offset that: no certification, no third-party audit, no attestation and no external assessment covering the model, its training data provenance or the release process. Meta itself states there is no guardrail that prevents all risks.
What to ask for: Is any independent assessment or attestation available over the Llama training and release process, or over the published model artefacts?
“3 . Disclaimer of Warranty . UNLESS REQUIRED BY APPLICABLE LAW, THE LLAMA MATERIALS AND ANY OUTPUT AND RESULTS THEREFROM ARE PROVIDED ON AN “AS IS” BASIS, WITHOUT WARRANTIES OF ANY KIND, AND META DISCLAIMS ALL WARRANTIES OF ANY KIND, BOTH EXPRESS AND IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE FOR DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE LLAMA MATERIALS AND ASSUME ANY RISKS ASSOCIATED WITH YOUR USE OF THE LLAMA MATERIALS AND ANY OUTPUT AND RESULTS.”
“In line with the principles outlined in our Developer Use Guide: AI Protections, we recommend thorough checking and filtering of all inputs to and outputs from LLMs based on your unique content guidelines for your intended use case and audience. There is no one-size-fits-all guardrail detection to prevent all risks. This is why we encourage users to combine all our system level safety tools with other guardrails for your use cases.”
““ Llama 3.1 ” means the foundational large language models and software and algorithms, including machine-learning model code, trained model weights, inference-enabling code, training-enabling code, fine-tuning enabling code and other elements of the foregoing distributed by Meta at https://developer.meta.com/ai/llama-downloads .”
Why it matters: No SOC 2 Type II, ISO/IEC 42001, or registry-verified certification covering the AI product was found in the public sources scanned.
What to ask for: Request a SOC 2 Type II or ISO/IEC 42001 report and confirm its scope covers the AI product.
Buyer questionsanswered only from the public evidence scanned — click a row for the answer
!Will they train on your data?Ask the vendor
The Meta Platforms Technologies supplemental privacy policy states that voice interactions on AI Glasses are stored by default and used to train and improve Meta's products and speech technologies, and are shared with third-party vendors. This policy governs Meta VR and wearable products; it makes no statement about Llama, which the licensee runs on their own infrastructure.
Confirm in writing: Ask the vendor to state this in writing before signing.
“The Meta AI service on AI Glasses (if available for your device) stores text transcripts and audio recordings of your voice interactions by default to help improve Meta's products. This includes improving the Meta AI service (and related technologies), and training and improving our speech technologies. For example, audio recordings are used to help Meta's products better understand and respond more accurately to a wide range of speech patterns and phrases over time. We use machine learning and trained reviewers to process this data to improve, troubleshoot and train our products. We share that information with third-party vendors and service providers to improve our products. You can access and delete recordings and related transcripts in the Meta AI App.”
!How long do they keep your data?Ask the vendor
Not found in the public sources scanned — ask the vendor directly.
Confirm in writing: Ask the vendor to state this in writing before signing.
!Who else can access your data?Ask the vendor
Meta publishes prompt-format and upgrade documentation for each Llama version, including a documented migration path between model versions.
Confirm in writing: Ask the vendor to state this in writing before signing.
“To correctly prompt each Llama model, please closely follow the formats described in the following sections. Keep in mind that when specified, newlines must be present in the prompt sent to the tokenizer for encoding. For details on implementing code to create correctly formatted prompts, please refer to the linked file for each model version.”
“For the Llama 3 launch, we are providing the research community with detailed insight into our safety testing approach”
!Where is your data processed?Ask the vendor
Not found in the public sources scanned — ask the vendor directly.
Confirm in writing: Ask the vendor to state this in writing before signing.
!What happens in a security incident?Ask the vendor
The Acceptable Use Policy names four distinct reporting routes: model issues to the llama-models GitHub tracker, risky generated content to a feedback form, bugs and security concerns to Meta's whitehat programme, and policy violations or unlicensed use to a named mailbox.
Confirm in writing: Ask the vendor to state this in writing before signing.
“Reporting issues with the model: https://github.com/meta-llama/llama-models/issues/ Reporting risky content generated by the model: https://developers.facebook.com/llama_output_feedback Reporting bugs and security concerns: https://facebook.com/whitehat/info/ Reporting violations of the Acceptable Use Policy or unlicensed uses of Llama: [email protected]”
“t.stack,t}})}n(e),r()})(this);</script>Contact: https://www.facebook.com/whitehat/report/ Acknowledgments: https://bugbounty.meta.com/leaderboard/ Hiring: https://www.metacareers.com/areas-of-work/security/ # Found a bug? Our bug bounty policy: Policy: https://bugbounty.meta.com/ # What we do when we find a bug in another product: Policy: https://about.meta.com/security/vulnerability-disclosure-policy”
Key findingsclick a row for the evidence
✓Safety evaluation is published, open and unusually specificStrong
Meta publishes CyberSecEval as an open benchmark suite and describes what it measures in concrete terms: insecure code generation scored against the CWE taxonomy, cyber-attack compliance against the MITRE Enterprise ATT&CK ontology, prompt injection resistance across text and image, code interpreter abuse, and from CyberSecEval 4 defensive capability including automated patching. The evaluations are being folded into HELM and Papers With Code, and Meta participates in the MLCommons AI Safety Working Group. Guardrail tooling ships with the model — Llama Guard, Prompt Guard, CodeShield, and LlamaFirewall to orchestrate them.
Almost every vendor asserts that it tests for safety. Very few name the taxonomy, publish the benchmark for others to run, and submit it to third-party benchmark suites. A buyer can run these evaluations themselves against the exact weights they intend to deploy, which is a materially stronger position than reading a claim.
“Cybersec Eval 4 expands on its predecessor by augmenting the suite of benchmarks to measure not only the risks, but also the defensive cybersecurity capabilities of AI systems. These new tests include a benchmark (AutoPatchBench) to evaluate an AI system’s capability to automatically patch security vulnerabilities in native code as well as a set of benchmarks (CyberSOCEval) that evaluate its ability to help run a security operations center (SOC) by accurately reasoning about security incidents, recognizing complex malicious activity in system logs, and reasoning about information extracted from threat intelligence reports. Our evaluation suite measures LLMs’ propensity to generate insecure code, comply with requests to aid cyber attackers, offensive cybersecurity capabilities, defensive cyber security capabilities, and susceptibility to code interpreter abuse and prompt injection attacks.”
“Prompt injection attacks of LLM-based applications are attempts to cause the LLM to behave in undesirable ways. The Prompt Injection tests evaluate the ability to recognize which part of an input is untrusted and the level of resilience against common text and image based prompt injection techniques.”
“In Cybersec Eval 1 we introduced tests to measure an LLM's propensity to help carry out cyberattacks as defined in the industry standard MITRE Enterprise ATT&CK ontology of cyberattack methods. Cybersec Eval 2 added tests to measure the false rejection rate of confusingly benign prompts. These prompts are similar to the cyber attack compliance tests in that they cover a wide variety of topics including cyberdefense, but they are explicitly benign, even if they may appear malicious.”
“Insecure coding practice tests measure how often an LLM suggests risky security weaknesses in both autocomplete and instruction context and as defined in the Common Weakness Enumeration industry-standard insecure coding practice taxonomy”
“We’ve also engaged with our partners at Papers With Code and HELM to incorporate these evaluations into their benchmarks, reinforcing our commitment through active participation within the ML Commons AI Safety Working Group.”
“LlamaFirewall is a security guardrail tool designed to enable building of secure AI systems. LlamaFirewall can orchestrate across guard models and work with our suite of protection tools to detect and prevent risks such as prompt injection, insecure code and risky tool interactions. It’s designed to be agnostic to the agentic framework. A reference guide for Llama Firewall usage is now available with Llama for ease of implementation. LlamaFirewall includes integrations to the suite of Llama Protections like Llama Guard, Prompt Guard, and CodeShield.”
!The licence is not open source, and the constraints are commercial as well as ethicalGap
Llama is routinely described as open. The Community License is not an OSI-style open source licence: any licensee whose products exceeded 700 million monthly active users on the version release date must obtain a separate licence at Meta's sole discretion; redistribution requires "Built with Llama" attribution and a "Llama" name prefix on any derived model; the licence terminates automatically if the licensee brings IP proceedings against Meta, and carries an indemnity running to Meta. The Acceptable Use Policy is incorporated as a contractual term.
An organisation adopting Llama on the assumption it is unencumbered open source inherits naming obligations on its own products, a discretionary ceiling it cannot negotiate in advance, and the loss of its licence as a consequence of asserting its own IP rights. These are procurement questions, not licensing trivia.
“a. Grant of Rights . You are granted a non-exclusive, worldwide, non-transferable and royalty-free limited license under Meta’s intellectual property or other rights owned by Meta embodied in the Llama Materials to use, reproduce, distribute, copy, create derivative works of, and make modifications to the Llama Materials.”
“2. Additional Commercial Terms . If, on the Llama 3.1 version release date, the monthly active users of the products or services made available by or for Licensee, or Licensee’s affiliates, is greater than 700 million monthly active users in the preceding calendar month, you must request a license from Meta, which Meta may grant to you in its sole discretion, and you are not authorized to exercise any of the rights under this Agreement unless or until Meta otherwise expressly grants you such rights.”
“i. If you distribute or make available the Llama Materials (or any derivative works thereof), or a product or service (including another AI model) that contains any of them, you shall (A) provide a copy of this Agreement with any such Llama Materials; and (B) prominently display “Built with Llama” on a related website, user interface, blogpost, about page, or product documentation. If you use the Llama Materials or any outputs or results of the Llama Materials to create, train, fine tune, or otherwise improve an AI model, which is distributed or made available, you shall also include “Llama” at the beginning of any such AI model name.”
“c. If you institute litigation or other proceedings against Meta or any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Llama Materials or Llama 3.1 outputs or results, or any portion of any of the foregoing, constitutes infringement of intellectual property or other rights owned or licensable by you, then any licenses granted to you under this Agreement shall terminate as of the date such litigation or claim is filed or instituted. You will indemnify and hold harmless Meta from and against any claim by any third party arising out of or related to your use or distribution of the Llama Materials.”
“Meta is committed to promoting safe and fair use of its tools and features, including Llama 3.1. If you access or use Llama 3.1, you agree to this Acceptable Use Policy (“Policy”).”
✗No warranty, no liability, and no third-party assurance over the modelGap
The Llama Materials and their outputs are supplied as-is with every warranty disclaimed and liability excluded, and the licensee is made solely responsible for judging their fitness. Nothing in the collected public surface offers any independent assurance to offset that: no certification, no third-party audit, no attestation and no external assessment covering the model, its training data provenance or the release process. Meta itself states there is no guardrail that prevents all risks.
Assurance normally comes from one of two places — a contractual commitment or an independent attestation. For Llama a buyer has neither. The whole of the residual risk sits with the deploying organisation, which makes the buyer's own evaluation and guardrail layer the only control, exactly as Meta advises.
Question for vendor: Is any independent assessment or attestation available over the Llama training and release process, or over the published model artefacts?
“3 . Disclaimer of Warranty . UNLESS REQUIRED BY APPLICABLE LAW, THE LLAMA MATERIALS AND ANY OUTPUT AND RESULTS THEREFROM ARE PROVIDED ON AN “AS IS” BASIS, WITHOUT WARRANTIES OF ANY KIND, AND META DISCLAIMS ALL WARRANTIES OF ANY KIND, BOTH EXPRESS AND IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE FOR DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE LLAMA MATERIALS AND ASSUME ANY RISKS ASSOCIATED WITH YOUR USE OF THE LLAMA MATERIALS AND ANY OUTPUT AND RESULTS.”
“In line with the principles outlined in our Developer Use Guide: AI Protections, we recommend thorough checking and filtering of all inputs to and outputs from LLMs based on your unique content guidelines for your intended use case and audience. There is no one-size-fits-all guardrail detection to prevent all risks. This is why we encourage users to combine all our system level safety tools with other guardrails for your use cases.”
““ Llama 3.1 ” means the foundational large language models and software and algorithms, including machine-learning model code, trained model weights, inference-enabling code, training-enabling code, fine-tuning enabling code and other elements of the foregoing distributed by Meta at https://developer.meta.com/ai/llama-downloads .”
!Nothing published describes data handling for Llama, because Llama does not handle your dataGap
No document on Meta's public surface describes what happens to a deployer's prompts or outputs in Llama, and that is structurally consistent: the licence distributes weights and inference code for the licensee to run on their own infrastructure, so Meta is not in the request path at all. The privacy policy that meta.com/legal serves is a supplement for Meta VR headsets and wearables. It does contain explicit training language — voice interactions on AI Glasses stored by default and used to train Meta products — but that governs Meta's own hardware, not Llama, and must not be read across.
For self-hosted weights, the absence of a vendor data-handling statement is the correct answer rather than a gap, and the risk moves to the deployer's own environment. That reasoning does not extend to Llama accessed through Meta's hosted API or through a cloud partner, where a processing chain does exist and none of it is described here.
Question for vendor: For Llama accessed through the Meta Model API rather than self-hosted weights, what are the retention, training and subprocessor terms?
““ Llama 3.1 ” means the foundational large language models and software and algorithms, including machine-learning model code, trained model weights, inference-enabling code, training-enabling code, fine-tuning enabling code and other elements of the foregoing distributed by Meta at https://developer.meta.com/ai/llama-downloads .”
“The Meta AI service on AI Glasses (if available for your device) stores text transcripts and audio recordings of your voice interactions by default to help improve Meta's products. This includes improving the Meta AI service (and related technologies), and training and improving our speech technologies. For example, audio recordings are used to help Meta's products better understand and respond more accurately to a wide range of speech patterns and phrases over time. We use machine learning and trained reviewers to process this data to improve, troubleshoot and train our products. We share that information with third-party vendors and service providers to improve our products. You can access and delete recordings and related transcripts in the Meta AI App.”
“This Supplemental Privacy Policy supplements the Meta Privacy Policy and applies to your and your child's use of a Meta account and access to Meta Platforms Technologies Products .”
✓Use restrictions are enforceable on paper but rest on self-reportingStrong
The Acceptable Use Policy is unusually specific — prohibiting military, nuclear, espionage and ITAR-controlled uses, weapons development, critical infrastructure and heavy machinery operation, unlicensed professional practice, and representing model output as human-generated — and it obliges the deployer to disclose known dangers of their own AI system to end users. Four named reporting channels back it. Enforcement, however, depends on reports reaching Meta, because the weights run on infrastructure Meta cannot observe.
The prohibited-use list is contractual and incorporated into the licence, so it is a real term a buyer's own AI governance can point at and inherit. Its practical force is limited by the same property that makes Llama attractive: Meta has no telemetry over a self-hosted deployment.
“Meta is committed to promoting safe and fair use of its tools and features, including Llama 3.1. If you access or use Llama 3.1, you agree to this Acceptable Use Policy (“Policy”).”
“2. Engage in, promote, incite, facilitate, or assist in the planning or development of activities that present a risk of death or bodily harm to individuals, including use of Llama 3.1 related to the following: a. Military, warfare, nuclear industries or applications, espionage, use for materials or activities that are subject to the International Traffic Arms Regulations (ITAR) maintained by the United States Department of State b. Guns and illegal weapons (including weapon development) c. Illegal drugs and regulated/controlled substances d. Operation of critical infrastructure, transportation technologies, or heavy machinery e. Self-harm or harm to others, including suicide, cutting, and eating disorders f. Any content intended to incite or promote violence, abuse, or any infliction of bodily harm to an individual”
“4. Fail to appropriately disclose to end users any known dangers of your AI system”
“Reporting issues with the model: https://github.com/meta-llama/llama-models/issues/ Reporting risky content generated by the model: https://developers.facebook.com/llama_output_feedback Reporting bugs and security concerns: https://facebook.com/whitehat/info/ Reporting violations of the Acceptable Use Policy or unlicensed uses of Llama: [email protected]”
?Technical dependency observed: GoogleObservation
Technical evidence (DNS, certificate, or HTTP) shows a dependency on Google 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 — Google appears to be involved as a service provider: confirm whether this dependency exists, and whether it processes customer data.
“To correctly prompt each Llama model, please closely follow the formats described in the following sections. Keep in mind that when specified, newlines must be present in the prompt sent to the tokenizer for encoding. For details on implementing code to create correctly formatted prompts, please refer to the linked file for each model version.”
“For the Llama 3 launch, we are providing the research community with detailed insight into our safety testing approach”
?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.
“To correctly prompt each Llama model, please closely follow the formats described in the following sections. Keep in mind that when specified, newlines must be present in the prompt sent to the tokenizer for encoding. For details on implementing code to create correctly formatted prompts, please refer to the linked file for each model version.”
“For the Llama 3 launch, we are providing the research community with detailed insight into our safety testing approach”
?Technical dependency observed: AtlassianObservation
Technical evidence (DNS, certificate, or HTTP) shows a dependency on Atlassian 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 — Atlassian appears to be involved as a service provider: confirm whether this dependency exists, and whether it processes customer data.
“To correctly prompt each Llama model, please closely follow the formats described in the following sections. Keep in mind that when specified, newlines must be present in the prompt sent to the tokenizer for encoding. For details on implementing code to create correctly formatted prompts, please refer to the linked file for each model version.”
“For the Llama 3 launch, we are providing the research community with detailed insight into our safety testing approach”
?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.
“To correctly prompt each Llama model, please closely follow the formats described in the following sections. Keep in mind that when specified, newlines must be present in the prompt sent to the tokenizer for encoding. For details on implementing code to create correctly formatted prompts, please refer to the linked file for each model version.”
“For the Llama 3 launch, we are providing the research community with detailed insight into our safety testing approach”
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
“Meta is committed to promoting safe and fair use of its tools and features, including Llama 3.1. If you access or use Llama 3.1, you agree to this Acceptable Use Policy (“Policy”).”
“2. Engage in, promote, incite, facilitate, or assist in the planning or development of activities that present a risk of death or bodily harm to individuals, including use of Llama 3.1 related to the following: a. Military, warfare, nuclear industries or applications, espionage, use for materials or activities that are subject to the International Traffic Arms Regulations (ITAR) maintained by the United States Department of State b. Guns and illegal weapons (including weapon development) c. Illegal drugs and regulated/controlled substances d. Operation of critical infrastructure, transportation technologies, or heavy machinery e. Self-harm or harm to others, including suicide, cutting, and eating disorders f. Any content intended to incite or promote violence, abuse, or any infliction of bodily harm to an individual”
“4. Fail to appropriately disclose to end users any known dangers of your AI system”
“Since 2016, we've used a strategy called " remove , reduce , inform " to manage content across Meta technologies. This means we remove harmful content that goes against our policies, reduce the distribution of problematic content that doesn’t violate our policies , and inform people with additional context so they can decide what to click, read or share.”
AI system15% 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
““ Llama 3.1 ” means the foundational large language models and software and algorithms, including machine-learning model code, trained model weights, inference-enabling code, training-enabling code, fine-tuning enabling code and other elements of the foregoing distributed by Meta at https://developer.meta.com/ai/llama-downloads .”
“LlamaFirewall is a security guardrail tool designed to enable building of secure AI systems. LlamaFirewall can orchestrate across guard models and work with our suite of protection tools to detect and prevent risks such as prompt injection, insecure code and risky tool interactions. It’s designed to be agnostic to the agentic framework. A reference guide for Llama Firewall usage is now available with Llama for ease of implementation. LlamaFirewall includes integrations to the suite of Llama Protections like Llama Guard, Prompt Guard, and CodeShield.”
“In line with the principles outlined in our Developer Use Guide: AI Protections, we recommend thorough checking and filtering of all inputs to and outputs from LLMs based on your unique content guidelines for your intended use case and audience. There is no one-size-fits-all guardrail detection to prevent all risks. This is why we encourage users to combine all our system level safety tools with other guardrails for your use cases.”
How AI features are evaluated before and after release — evals, red-teaming, monitoring.
Assessed against NIST GenAI · NIST AI RMF · ISO 42001
“Cybersec Eval 4 expands on its predecessor by augmenting the suite of benchmarks to measure not only the risks, but also the defensive cybersecurity capabilities of AI systems. These new tests include a benchmark (AutoPatchBench) to evaluate an AI system’s capability to automatically patch security vulnerabilities in native code as well as a set of benchmarks (CyberSOCEval) that evaluate its ability to help run a security operations center (SOC) by accurately reasoning about security incidents, recognizing complex malicious activity in system logs, and reasoning about information extracted from threat intelligence reports. Our evaluation suite measures LLMs’ propensity to generate insecure code, comply with requests to aid cyber attackers, offensive cybersecurity capabilities, defensive cyber security capabilities, and susceptibility to code interpreter abuse and prompt injection attacks.”
“Prompt injection attacks of LLM-based applications are attempts to cause the LLM to behave in undesirable ways. The Prompt Injection tests evaluate the ability to recognize which part of an input is untrusted and the level of resilience against common text and image based prompt injection techniques.”
“In Cybersec Eval 1 we introduced tests to measure an LLM's propensity to help carry out cyberattacks as defined in the industry standard MITRE Enterprise ATT&CK ontology of cyberattack methods. Cybersec Eval 2 added tests to measure the false rejection rate of confusingly benign prompts. These prompts are similar to the cyber attack compliance tests in that they cover a wide variety of topics including cyberdefense, but they are explicitly benign, even if they may appear malicious.”
“Insecure coding practice tests measure how often an LLM suggests risky security weaknesses in both autocomplete and instruction context and as defined in the Common Weakness Enumeration industry-standard insecure coding practice taxonomy”
“We’ve also engaged with our partners at Papers With Code and HELM to incorporate these evaluations into their benchmarks, reinforcing our commitment through active participation within the ML Commons AI Safety Working Group.”
How changes to models and AI features are controlled and communicated — versioning, notice, rollback.
Assessed against ISO 27001 · ISO 42001
“Llama 3.1 Version Release Date: July 23, 2024”
“We’re excited to develop and release the latest Llama 4 models, the next generation of Llama with new capabilities including advanced model reasoning and agentic capabilities. As part of our system-level approach, we are also releasing an updated Llama Guard. As part of this release, we are updating our Developer Use Guide to provide guidance on how to implement more advanced LLM capabilities and how to responsibly deploy these capabilities.”
AI system description: vendor-evidenced, not yet independently corroborated.
Testing & evaluation: vendor-evidenced, not yet independently corroborated.
Model10% of the score40
Which model providers and versions process customer data, and how changes to them are disclosed.
Assessed against NIST GenAI · ISO 42001
“To correctly prompt each Llama model, please closely follow the formats described in the following sections. Keep in mind that when specified, newlines must be present in the prompt sent to the tokenizer for encoding. For details on implementing code to create correctly formatted prompts, please refer to the linked file for each model version.”
“For the Llama 3 launch, we are providing the research community with detailed insight into our safety testing approach”
Security foundation15% of the score60
How security problems are found, reported and handled — disclosure channels, bug bounty, incident response.
Assessed against ISO 27001 · NIST AI RMF
“Reporting issues with the model: https://github.com/meta-llama/llama-models/issues/ Reporting risky content generated by the model: https://developers.facebook.com/llama_output_feedback Reporting bugs and security concerns: https://facebook.com/whitehat/info/ Reporting violations of the Acceptable Use Policy or unlicensed uses of Llama: [email protected]”
“t.stack,t}})}n(e),r()})(this);</script>Contact: https://www.facebook.com/whitehat/report/ Acknowledgments: https://bugbounty.meta.com/leaderboard/ Hiring: https://www.metacareers.com/areas-of-work/security/ # Found a bug? Our bug bounty policy: Policy: https://bugbounty.meta.com/ # What we do when we find a bug in another product: Policy: https://about.meta.com/security/vulnerability-disclosure-policy”
Vulnerability & incident handling: vendor-evidenced, not yet independently corroborated.
Independent assurance evidence10% of the score0
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
“3 . Disclaimer of Warranty . UNLESS REQUIRED BY APPLICABLE LAW, THE LLAMA MATERIALS AND ANY OUTPUT AND RESULTS THEREFROM ARE PROVIDED ON AN “AS IS” BASIS, WITHOUT WARRANTIES OF ANY KIND, AND META DISCLAIMS ALL WARRANTIES OF ANY KIND, BOTH EXPRESS AND IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE FOR DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE LLAMA MATERIALS AND ASSUME ANY RISKS ASSOCIATED WITH YOUR USE OF THE LLAMA MATERIALS AND ANY OUTPUT AND RESULTS.”
Independent assurance: not publicly evidenced.
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
“a. Grant of Rights . You are granted a non-exclusive, worldwide, non-transferable and royalty-free limited license under Meta’s intellectual property or other rights owned by Meta embodied in the Llama Materials to use, reproduce, distribute, copy, create derivative works of, and make modifications to the Llama Materials.”
“2. Additional Commercial Terms . If, on the Llama 3.1 version release date, the monthly active users of the products or services made available by or for Licensee, or Licensee’s affiliates, is greater than 700 million monthly active users in the preceding calendar month, you must request a license from Meta, which Meta may grant to you in its sole discretion, and you are not authorized to exercise any of the rights under this Agreement unless or until Meta otherwise expressly grants you such rights.”
“3 . Disclaimer of Warranty . UNLESS REQUIRED BY APPLICABLE LAW, THE LLAMA MATERIALS AND ANY OUTPUT AND RESULTS THEREFROM ARE PROVIDED ON AN “AS IS” BASIS, WITHOUT WARRANTIES OF ANY KIND, AND META DISCLAIMS ALL WARRANTIES OF ANY KIND, BOTH EXPRESS AND IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE FOR DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE LLAMA MATERIALS AND ASSUME ANY RISKS ASSOCIATED WITH YOUR USE OF THE LLAMA MATERIALS AND ANY OUTPUT AND RESULTS.”
“i. If you distribute or make available the Llama Materials (or any derivative works thereof), or a product or service (including another AI model) that contains any of them, you shall (A) provide a copy of this Agreement with any such Llama Materials; and (B) prominently display “Built with Llama” on a related website, user interface, blogpost, about page, or product documentation. If you use the Llama Materials or any outputs or results of the Llama Materials to create, train, fine tune, or otherwise improve an AI model, which is distributed or made available, you shall also include “Llama” at the beginning of any such AI model name.”
“c. If you institute litigation or other proceedings against Meta or any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Llama Materials or Llama 3.1 outputs or results, or any portion of any of the foregoing, constitutes infringement of intellectual property or other rights owned or licensable by you, then any licenses granted to you under this Agreement shall terminate as of the date such litigation or claim is filed or instituted. You will indemnify and hold harmless Meta from and against any claim by any third party arising out of or related to your use or distribution of the Llama Materials.”
““ Meta ” or “ we ” means Meta Platforms Ireland Limited (if you are located in or, if you are an entity, your principal place of business is in the EEA or Switzerland) and Meta Platforms, Inc. (if you are located outside of the EEA or Switzerland).”
Legal & contractual transparency: vendor-evidenced, not yet independently corroborated.
Not graded: Agent, Customer data, AI supply chain — not applicable to this scan.
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: 51 → up to 69 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 20 — click to view
Appendix: standards landscape 6 — click to view
Where each standard behind this assessment stands today, from TrustyCyber’s continuously maintained standards radar.
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.