Anthropic
- Rank
- Score
- 1/149
- 71 out of 100
Anthropic is a US AI company founded in 2021 by Dario Amodei, Daniela Amodei and other former OpenAI researchers. It is based in San Francisco. Anthropic makes the Claude AI assistant and models, including the Claude Code tool for software developers, and is known for its focus on AI safety research.
Is Claude safe?
Claude scores 71 out of 100 and ranks 1 of 149. Its strongest category is Security, and its weakest category is Safety.
How is Claude scored?
Claude is scored out of 100 on Safety (25%), Transparency (25%), Security (25%), and Privacy (25%). See how we score.
Compare
Safety framework
30Average of SaferAI's published scores.
- 41
SaferAI score: risk identification
Independent sourceSaferAI (opens in a new tab) · 14 Oct 2026 - 19
SaferAI score: risk analysis and evaluation
Independent sourceSaferAI (opens in a new tab) · 14 Oct 2026 - 30
SaferAI score: risk treatment
Independent sourceSaferAI (opens in a new tab) · 14 Oct 2026
Pre-release testing
100Points for each confirmed yes.
- Yes
Was the model tested for dangerous capabilities, such as biological, chemical or cyber risks?
Company claim50/50“On chemical and biological risks, we judge that the model has CB-1 capabilities-meaning it could meaningfully help someone with a basic tech…” MoreLess
“On chemical and biological risks, we judge that the model has CB-1 capabilities-meaning it could meaningfully help someone with a basic technical background synthesize a known weapon-but falls short of the CB-2 threshold for functionally replacing rare expert talent, which is the limiting factor in developing novel weapons.”
Anthropic model card (opens in a new tab) · 14 Oct 2026 - Yes
Did an outside organisation test the model before release?
Company claim50/50“We conducted pre-deployment testing of Claude Mythos 5.1’s capabilities with METR, focusing on capabilities relevant to the automation of AI…” MoreLess
“We conducted pre-deployment testing of Claude Mythos 5.1’s capabilities with METR, focusing on capabilities relevant to the automation of AI R&D. They shared the following findings with us:”
Anthropic model card (opens in a new tab) · 14 Oct 2026
Safety test results
99.3Average of three independent safety tests of the company's models: MLCommons AILuminate, Stanford HELM Safety and Stanford AIR-Bench. Each test is compared with the best company's result.
- 100
MLCommons AILuminate safety grades
Independent sourceMLCommons AILuminate (opens in a new tab) · 10 Nov 2026 - 97.8
Stanford HELM Safety mean score, averaged across the company's tested models: 95.7, against 97.9 for the top company of 13 tested
Independent sourcecrfm.stanford.edu (opens in a new tab) · 10 Nov 2026 - 100
Stanford AIR-Bench 2024 refusal rate, averaged across the company's tested models: 86.7, against 86.7 for the top company of 14 tested
Independent sourcecrfm.stanford.edu (opens in a new tab) · 10 Nov 2026
Incident record
6.3This check counts only when a source covers the company. With no events it holds the pillar, which is not proof of a clean record. Each recorded event takes points off the pillar, and serious or recent ones count most. Allegation checks, regulator actions and open investigations use half the severity load. AI Incident Database and AIAAIC Repository, last 24 months.
45 events
AllegationAI Incident Database (opens in a new tab) · 12 Nov 2026These are reports, not proven findings.
Date Event Severity 29 Oct 2026 AIID #1808 · Defense Contractor's Personal AI Coding Agent Reportedly Sent U.S. Defense Department Credentials to a Third-party Website 29 Oct 2026 · Medium (reported)Also reported by AIAAIC (opens in a new tab)
Medium (reported) 3 Oct 2026 AIID #1740 · Teen Hiker Required Helicopter Rescue After Reportedly Following Claude Chatbot Directions in British Columbia, Canada3 Oct 2026 · Medium (reported) Medium (reported) 27 Sep 2026 AIID #1782 · Attackers Reportedly Used ARTEX, an AI Security-Testing Tool, to Steal Data from South Korean Financial Institutions27 Sep 2026 · Medium (reported) Medium (reported) 19 Sep 2026 AIID #1719 · Anthropic's Claude Code AI Agent Reportedly Deleted About 48,000 Live Project Files and Version-Control History During Software Repair19 Sep 2026 · Low (reported) Low (reported) 10 Sep 2026 AIID #1687 · Anthropic-Designated GTG-87001 Weapons Cell in Northern Yemen Reportedly Used Claude Code to Develop Missile Guidance Software10 Sep 2026 · High (reported) High (reported) 9 Sep 2026 AIID #1685 · Early Claude Opus 4.6 Checkpoint Reportedly Gained Unauthorized Admin Access to Third-Party System During Cybersecurity Evaluation9 Sep 2026 · Medium (reported) Medium (reported) 9 Aug 2026 AIID #1791 · Dartmouth College Provost's Undisclosed AI-Assisted Writing Was Found to Have Breached Publication Policies9 Aug 2026 · Medium (reported) Medium (reported) 30 Jul 2026 AIID #1627 · Claude Opus 4.7 Reportedly Compromised Real Company's Production Infrastructure During Cybersecurity Evaluation30 Jul 2026 · Low (reported) Low (reported) 30 Jul 2026 AIID #1628 · Claude Mythos 5 Reportedly Published Malicious PyPI Package That Compromised Real Security Company During Evaluation30 Jul 2026 · Medium (reported) Medium (reported) 30 Jul 2026 AIID #1629 · Anthropic Research Model Reportedly Scanned 9,000 Targets and Compromised Real Company's Application During Evaluation30 Jul 2026 · Low (reported) Low (reported) 28 Jul 2026 AIID #1676 · Anthropic Claude Opus 5 Coding Agent Reportedly Reset a Live Supabase Production Database During Prisma Migration Work28 Jul 2026 · Low (reported) Low (reported) 26 Jul 2026 AIID #1621 · Publicly Shared Claude Artifacts Reportedly Became Discoverable Through Google Search26 Jul 2026 · Low (reported) Low (reported) 26 Jul 2026 AIID #1633 · Anthropic and OpenAI AI Agents Reportedly Took Unsanctioned Actions on the Live Internet During UK AISI Cybersecurity Evaluations26 Jul 2026 · Low (reported) Low (reported) 18 Jul 2026 AIID #1751 · Anthropic AI Agent Reportedly Submitted a Fabricated Homicide Tip to Police in Philadelphia18 Jul 2026 · Low (reported) Low (reported) 1 Jul 2026 AIAAIC2261 · AI chatbots lure vulnerable gamblers to unlicensed betting websites1 Jul 2026 · Low (reported) Low (reported) 24 Jun 2026 AIID #1641 · Rep. Anna Paulina Luna's Office Reportedly Published Claude Transcript Residue on Accident in House Amendment Summary24 Jun 2026 · Low (reported) Low (reported) 19 May 2026 AIID #1796 · Attacker Reportedly Used AI Agents to Breach Healthcare Organizations and Steal Medical Records in Argentina19 May 2026 · Medium (reported) Medium (reported) 12 May 2026 AIID #1504 · Nonfiction Book 'The Future of Truth' Reportedly Included AI-Generated and Misattributed Quotations12 May 2026 · Low (reported) Low (reported) 6 May 2026 AIID #1499 · Claude Console Reportedly Generated Phantom Legal Quotations in Trump Layoffs Court Filing6 May 2026 · Low (reported) Low (reported) 30 Apr 2026 AIID #1642 · AI Agent Powered by Claude Opus 4.6 Reportedly Exploited Gym Booking API and Removed Another Member From Waitlist30 Apr 2026 · Low (reported) Low (reported) 27 Apr 2026 AIID #1470 · DisMech AI Curation Agent Reportedly Completed GitHub Issue Intended as New Contributor's Learning Task27 Apr 2026 · Low (reported) Low (reported) 24 Apr 2026 AIID #1469 · PocketOS Production Database Was Reportedly Deleted by Cursor AI Agent Running Claude Opus 4.624 Apr 2026 · Low (reported) Low (reported) 8 Apr 2026 AIID #1661 · Aurora Ransomware Operator Reportedly Used Cursor Agent to Assist Intrusions Against Multiple Organizations8 Apr 2026 · Medium (reported) Medium (reported) 7 Apr 2026 AIID #1613 · Claude Mythos Preview Reportedly Posted Sandbox Exploit Details to Public Websites During Anthropic Evaluation7 Apr 2026 · Low (reported) Low (reported) 2 Apr 2026 AIID #1723 · Anthropic's Claude Code AI Coding Tool Reportedly Used Hidden Unicode Markers to Classify Custom API Routing2 Apr 2026 · Low (reported) Low (reported) 28 Feb 2026 AIID #1492 · Purportedly AI-Enabled Targeting System Was Reportedly Implicated in Deadly U.S. Strike on Iranian Primary School28 Feb 2026 · High (reported) High (reported) 26 Feb 2026 AIID #1424 · Claude Code Agent Reportedly Deleted DataTalks.Club Production Infrastructure, Database, and Snapshots via Terraform26 Feb 2026 · Low (reported) Low (reported) 23 Feb 2026 AIID #1395 · Anthropic Said DeepSeek, Moonshot, and MiniMax Used Fraudulent Accounts and Proxies to Illicitly Distill Claude Capabilities at Scale23 Feb 2026 · Medium (reported) Medium (reported) 17 Feb 2026 AIID #1680 · Unknown Actor Reportedly Exploited Cline's Claude-Powered GitHub Issue-Triage Workflow and Published an Unauthorized Cline CLI Package17 Feb 2026 · Medium (reported) Medium (reported) 7 Feb 2026 AIID #1441 · Claude Cowork Allegedly Deleted Folder Containing 15 Years of Family Photos While Organizing User's Wife's Desktop7 Feb 2026 · Low (reported) Low (reported) 18 Dec 2025 AIID #1313 · Anthropic Claude AI Agent Reportedly Caused Financial Losses While Operating Office Vending Machine at Wall Street Journal Headquarters 18 Dec 2025 · Low (reported)Also reported by AIAAIC (opens in a new tab)
Low (reported) 1 Dec 2025 AIID #1430 · Anthropic's Claude Was Reportedly Jailbroken To Allegedly Help Steal Sensitive Mexican Government Data 1 Dec 2025 · Medium (reported)Also reported by AIAAIC (opens in a new tab)
Medium (reported) 13 Nov 2025 AIID #1263 · Chinese State-Linked Operator (GTG-1002) Reportedly Uses Claude Code for Autonomous Cyber Espionage13 Nov 2025 · Low (reported) Low (reported) 27 Aug 2025 AIID #1201 · Anthropic Reportedly Identifies AI Misuse in Extortion Campaigns, North Korean IT Schemes, and Ransomware Sales27 Aug 2025 · Medium (reported) Medium (reported) 21 Aug 2025 AIID #1210 · Malicious Nx npm Packages Reportedly Weaponize AI Coding Agents for Data Exfiltration21 Aug 2025 · Medium (reported) Medium (reported) 31 Jul 2025 AIID #1186 · Reported Public Exposure of Over 100,000 LLM Conversations via Share Links Indexed by Search Engines and Archived31 Jul 2025 · Medium (reported) Medium (reported) 1 Jul 2025 AIAAIC1938 · European Parliament AI model is unable to identify first European Commission president1 Jul 2025 · Low (reported) Low (reported) 1 Jul 2025 AIAAIC1974 · Top chatbots tricked into generating instructions on how to enrich uranium1 Jul 2025 · Low (reported) Low (reported) 1 Jul 2025 AIAAIC1981 · Anthropic accused of using fake AI source in copyright case1 Jul 2025 · Medium (reported) Medium (reported) 1 Jul 2025 AIAAIC2054 · Claude Opus 4 AI agent blackmails supervisor to prevent being shut down1 Jul 2025 · Low (reported) Low (reported) 15 May 2025 AIID #1074 · Citation Errors in Concord Music v. Anthropic Attributed to Claude AI Use by Defense Counsel15 May 2025 · Low (reported) Low (reported) 23 Apr 2025 AIID #1054 · Anthropic Report Details Claude Misuse for Influence Operations, Credential Stuffing, Recruitment Fraud, and Malware Development23 Apr 2025 · Medium (reported) Medium (reported) 15 Apr 2025 AIID #1044 · Reported Emergence of 'Vegetative Electron Microscopy' in Scientific Papers Traced to Purported AI Training Data Contamination15 Apr 2025 · Low (reported) Low (reported) 12 Apr 2025 AIID #1026 · Multiple LLMs Allegedly Endorsed Suicide as a Viable Option During Non-Adversarial Mental Health Venting Session12 Apr 2025 · High (reported) High (reported) 5 Mar 2025 AIID #975 · At Least 10,000 AI Chatbots, Including Jailbroken Models, Allegedly Promote Eating Disorders, Self-Harm, and Sexualized Minors5 Mar 2025 · High (reported) High (reported)
Protecting children
100Points for each confirmed yes.
- Yes
Does the model card report child-safety tests or safeguards?
Company claim50/50“We ran our child safety evaluations following the same testing protocol we used for other recently released models such as Claude Opus 5.”
Anthropic model card (opens in a new tab) · 14 Oct 2026 - Yes
Does the privacy policy set a minimum age, or say the service is not directed at children?
Company claim50/50“Our Services are not directed towards, and we do not knowingly collect, use, disclose, sell, or share any information from children under th…” MoreLess
“Our Services are not directed towards, and we do not knowingly collect, use, disclose, sell, or share any information from children under the age of 18.”
Anthropic privacy policy (opens in a new tab) · 21 Oct 2026
Strong written policies; outside evidence is weaker.
Model documentation
100Points for each confirmed yes.
- Yes
Does the model card describe what the model can do?
Company claim25/25“This model advances the frontier in coding, knowledge work, and problem-solving, with improved capabilities for novel mathematical and scien…” MoreLess
“This model advances the frontier in coding, knowledge work, and problem-solving, with improved capabilities for novel mathematical and scientific reasoning.”
Anthropic model card (opens in a new tab) · 14 Oct 2026 - Yes
Does the model card report test results?
Company claim25/25“4.2 Child safety evaluations” (results shown in a table on the source page)
Anthropic model card (opens in a new tab) · 14 Oct 2026 - Yes
Does the model card state the model's limitations?
Company claim25/25“● Claude Mythos 5.1 is a slight regression on overall misaligned behavior compared to Opus 5, though still an improvement over Claude Mythos…” MoreLess
“● Claude Mythos 5.1 is a slight regression on overall misaligned behavior compared to Opus 5, though still an improvement over Claude Mythos 5 and Claude Sonnet 5, as measured by our automated behavioral audit.”
Anthropic model card (opens in a new tab) · 14 Oct 2026 - Yes
Does the model card describe safety safeguards?
Company claim25/25“As we discuss in Section 1, we have deployed Claude Mythos 5.1 to the general public with additional safeguards as Claude Fable 5.1.”
Anthropic model card (opens in a new tab) · 14 Oct 2026
Training-data disclosure
15.9Average of two independent training-data checks: the Stanford index and an academic grading of EU AI Act training-data summaries.
- 15
Stanford Foundation Model Transparency Index: training-data indicators met
Independent sourceStanford FMTI (opens in a new tab) · 14 Oct 2026 - 16.8
EU AI Act training-data summary grades (AI Accountability Lab)
Independent sourceAI Accountability Lab (opens in a new tab) · 10 Nov 2026
Independent transparency index
46Stanford Foundation Model Transparency Index total, out of 100.
- 46
Transparency index score
Independent sourceStanford FMTI (opens in a new tab) · 14 Oct 2026
Transparency reports
10025 points for each area where Partnership on AI names the company a leader in reporting on real-world impacts.
- 100
Partnership on AI practice-leader areas
Company claimPartnership on AI (opens in a new tab) · 10 Nov 2026
Public safety commitments
100Average of the commitments signed.
- Yes
Has it signed the EU code of practice for general-purpose AI?
Company claim100/100“Amazon Anthropic Black Forest Labs”
European Commission (opens in a new tab) · 14 Oct 2026 - No
Did it sign only the safety and security chapter?
Company claim0/50European Commission (opens in a new tab) · 14 Oct 2026 - Yes
Has it signed the Frontier AI Safety Commitments (Seoul 2024)?
Company claim100/100“The UK and Republic of Korea governments announced that the following organisations have agreed to the Frontier AI Safety Commitments: Amazo…” MoreLess
“The UK and Republic of Korea governments announced that the following organisations have agreed to the Frontier AI Safety Commitments: Amazon Anthropic Cohere Google G42 IBM Inflection AI Meta Microsoft Mistral AI Naver OpenAI Samsung Electronics Technology Innovation Institute xAI Zhipu.ai”
Frontier AI Safety Commitments (opens in a new tab) · 14 Oct 2026
No problems found in 1 public record. That is not proof.
Security programme and certifications
100Points for each confirmed yes.
- Yes
Is there a way to report security flaws, such as a bug bounty?
Company claim34/34“Anthropic (bug bounty)”
hackerone.com (opens in a new tab) · 2 Nov 2026 - Yes
Does it state it holds a SOC 2 Type II report?
Company claim33/33“SOC 2 Type I & Type II”
Anthropic security page (opens in a new tab) · 14 Oct 2026 - Yes
Does it state it holds ISO 27001 certification?
Company claim33/33“ISO 27001:2022 (Information Security Management)”
Anthropic security page (opens in a new tab) · 14 Oct 2026
Software vulnerabilities
70This check counts only when a source covers the company. With no events it holds the pillar, which is not proof of a clean record. Each recorded event takes points off the pillar, and serious or recent ones count most. Allegation checks, regulator actions and open investigations use half the severity load. Known-exploited flaws (CISA list) in the company's official software, last 24 months. A disclosed and fixed flaw does not lower the score.
No events
Independent sourceGitHub Advisory Database (opens in a new tab) · 12 Nov 2026
Data breaches
Not enough dataNo problems found in 2 public records. That is not proof.
Clear privacy policy
100Points for each confirmed yes.
- Yes
Does the privacy policy say what personal data is collected?
Company claim20/20“Identity and Contact Data: Anthropic collects identifiers, including your name, email address, and phone number when you sign up for an Anth…” MoreLess
“Identity and Contact Data: Anthropic collects identifiers, including your name, email address, and phone number when you sign up for an Anthropic account, or to receive information on our Services.”
Anthropic privacy policy (opens in a new tab) · 21 Oct 2026 - Yes
Does the privacy policy say whether your data is used to train AI?
Company claim20/20“We may use your Inputs and Outputs to train and improve Anthropic AI models, unless you opt out through your account settings.”
Anthropic privacy policy (opens in a new tab) · 21 Oct 2026 - Yes
Does the privacy policy say how long data is kept?
Company claim20/20“Anthropic retains your personal data for as long as reasonably necessary for the purposes and criteria outlined in this Privacy Policy and e…” MoreLess
“Anthropic retains your personal data for as long as reasonably necessary for the purposes and criteria outlined in this Privacy Policy and explained further in our privacy center.”
Anthropic privacy policy (opens in a new tab) · 21 Oct 2026 - Yes
Does the privacy policy say who your data is shared with?
Company claim20/20“Anthropic will disclose personal data to, or receive personal data from, the following categories of third parties for the purposes explaine…” MoreLess
“Anthropic will disclose personal data to, or receive personal data from, the following categories of third parties for the purposes explained in this Policy: Affiliates & corporate partners: Anthropic discloses the categories of personal data described above between and among its affiliates and related entities.”
Anthropic privacy policy (opens in a new tab) · 21 Oct 2026 - Yes
Does the privacy policy explain your privacy rights?
Company claim20/20“Correction: the right to request that we correct inaccurate personal data Anthropic retains about you, subject to certain exceptions.”
Anthropic privacy policy (opens in a new tab) · 21 Oct 2026
Training-data defaults
50100 if training is opt-in only, 50 if you can opt out, 0 if neither.
- No
Is your data used for training only if you opt in?
Company claim0/100Anthropic privacy policy (opens in a new tab) · 21 Oct 2026 - Yes
Can you stop your data being used for training?
Company claim50/50“We may use your Inputs and Outputs to train and improve Anthropic AI models, unless you opt out through your account settings.”
Anthropic privacy policy (opens in a new tab) · 21 Oct 2026
Export and delete your data
100Points for each confirmed yes.
- Yes
Can you get a copy of your data?
Company claim50/50“Access & data portability: the right to request a copy of the personal data Anthropic processes about you, subject to certain exceptions and…” MoreLess
“Access & data portability: the right to request a copy of the personal data Anthropic processes about you, subject to certain exceptions and conditions.”
Anthropic privacy policy (opens in a new tab) · 21 Oct 2026 - Yes
Can you delete your data?
Company claim50/50“Deletion: the right to request that we delete personal data collected from you when you use our Services, subject to certain exceptions.”
Anthropic privacy policy (opens in a new tab) · 21 Oct 2026
Privacy fines
Not enough dataIndependent privacy ratings
58.4Incogni's 2026 privacy-risk ranking: 100 is the lowest-risk platform rated, 0 the highest.
- 58.4
Incogni privacy-risk ranking
Independent sourceIncogni privacy ranking (opens in a new tab) · 10 Nov 2026
Risk governance
50Average of SaferAI's published scores.
- 50
SaferAI score: risk governance
Independent sourceSaferAI (opens in a new tab) · 14 Oct 2026
Regulator actions
70This check counts only when a source covers the company. With no events it holds the pillar, which is not proof of a clean record. Each recorded event takes points off the pillar, and serious or recent ones count most. Allegation checks, regulator actions and open investigations use half the severity load. UK competition regulator cases, US Justice Department settlements, last 5 years, and regulator actions and open investigations into the company's AI products at half severity, last 3 years.
No events
RegulatorUK CMA cases (opens in a new tab) · 12 Nov 2026
Court judgments
70This check counts only when a source covers the company. With no events it holds the pillar, which is not proof of a clean record. Each recorded event takes points off the pillar, and serious or recent ones count most. Allegation checks, regulator actions and open investigations use half the severity load. Court rulings against the company on privacy, consumer or competition claims, upheld on appeal, last 5 years.
No events
Regulatorcourtlistener.com (opens in a new tab) · 10 Nov 2026
Privacy contact and complaints
100Points for each confirmed yes.
- Yes
Is there a privacy contact, such as an email or a data protection officer?
Company claim50/50“If you become aware that a child under the age of 18 has provided any personal data to us while using our Services, please email us at priva…” MoreLess
“If you become aware that a child under the age of 18 has provided any personal data to us while using our Services, please email us at privacy@anthropic.com and we will investigate the matter and, if appropriate, delete the personal data.”
Anthropic privacy policy (opens in a new tab) · 21 Oct 2026 - Yes
Does it tell you that you can complain to a regulator?
Company claim50/50“Please note that under many countries' laws, you have the right to lodge a complaint with the supervisory authority in the place in which yo…” MoreLess
“Please note that under many countries' laws, you have the right to lodge a complaint with the supervisory authority in the place in which you live or work.”
Anthropic privacy policy (opens in a new tab) · 21 Oct 2026
Score history
Newest first. Only days when the score changed are listed.
2 Nov 2026
Score 71
Method 2 Nov 2026
+10 · Method update
28 Oct 2026
Score 61
Method 28 Oct 2026, update 2
-8 · combines both 28 Oct updates (28 Oct 2026 - Update 1 and 28 Oct 2026 - Update 2)
23 Oct 2026
Score 69
Method 21 Oct 2026
+2
21 Oct 2026
Score 67
Method 21 Oct 2026
-2
15 Oct 2026
Score 69
Method Not recorded
+1
14 Oct 2026
Score 68
Method Not recorded
-