--- title: Anthropic ran 133 million contractor chats with its... description: Anthropic's second Risk Report reveals that bioweapon-blocking classifiers were disabled on its human feedback platforms for eleven months (May 2025 to April... canonical: https://daily.dev/posts/anthropic-ran-133-million-contractor-chats-with-its-bioweapon-filters-off-ulyjzriyl twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Anthropic ran 133 million contractor chats with its bioweapon filters off | daily.dev og:description: Anthropic's second Risk Report reveals that bioweapon-blocking classifiers were disabled on its human feedback platforms for eleven months (May 2025 to April... og:url: https://daily.dev/posts/anthropic-ran-133-million-contractor-chats-with-its-bioweapon-filters-off-ulyjzriyl og:image: https://api.daily.dev/og/posts/ulYjzriyL.png og:image:alt: Anthropic ran 133 million contractor chats with its bioweapon filters off og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Anthropic ran 133 million contractor chats with its bioweapon filters off **[The Next Web](https://daily.dev/sources/tnw)** · 7 min read · 1 upvotes · 0 comments ## Summary Anthropic's second Risk Report reveals that bioweapon-blocking classifiers were disabled on its human feedback platforms for eleven months (May 2025 to April 2026), covering roughly 133 million contractor exchanges from about 50,000 people whose vetting Anthropic outsourced to vendors, some of which lacked screening capable of stopping serious threat actors. An internal-only flag silently disabled both the blocking and the logging, so incidents could not be reviewed after the fact. A review using Claude Sonnet 5 flagged 1,197 high-risk transcripts; manual review found no clear misuse but flagged some dual-use conversations, and Anthropic now says it believes similar undiscovered issues are more likely. The company also retroactively downgraded its February safety verdict from 'very low' to 'low' risk, disclosed a second incident where contractors misused a leaked API key to access an unreleased model called Mythos Preview for about two weeks, and included a self-critical review of the report written by a Claude instance, which flagged sections as overly reassuring and criticized a fully redacted incident. Anthropic also disclosed an unreleased internal model, Model 2, and rewrote its novel-weapons risk threshold. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why did Anthropic disable its bioweapon safety classifiers for contractors between May 2025 and April 2026? An internal flag meant only for internal use accidentally switched off both the blocking classifiers and the logging mechanism on Anthropic's human feedback platforms, so flagged harmful content was never recorded or sent to review. Roughly 50,000 contractors generated about 133 million exchanges during this period, vetted by outside vendors that in many cases could not screen out even CB-1 level threat actors. _Teams tracking AI vendor safety incidents can follow disclosures like this one on daily.dev._ ### What did Anthropic find when it reviewed contractor conversations for bioweapon-related risk after discovering the classifier outage? Claude Sonnet 5 scanned every human turn sent during the affected period and flagged 1,197 transcripts as high risk; 757 came from Anthropic's own teams and all but 62 of the remainder were deliberate red-teaming exercises. Staff manually reviewed the 62 plus 30 random red-team transcripts and found no clearly concerning misuse, though a handful of potentially dual-use conversations surfaced. _Following how AI labs audit their own safety failures helps teams evaluating which providers to trust on daily.dev._ ### What happened when contractors exploited a flaw to access Anthropic's Mythos Preview model outside their assigned work? A few contractors at data-labelling vendors exploited a flaw in April 2026 to obtain an API key and used models, including the capable Mythos Preview model, outside their assigned scope. Mythos Preview sat exposed for roughly two weeks without blocking biological classifiers before Anthropic contained the access within 90 minutes of learning about it, with no model weights, customer data, or core networks compromised. _daily.dev helps engineers weighing AI vendor risk stay current on incidents like this contractor access breach._ ## Similar posts on daily.dev - [Unauthorized users gained access to Anthropic’s restricted Mythos AI model](https://daily.dev/posts/unauthorized-users-gained-access-to-anthropic-s-restricted-mythos-ai-model-zrqbk3xot) · The Next Web · 0 upvotes · 0 comments - [Unauthorized group has gained access to Anthropic’s exclusive cyber tool Mythos, report claims](https://daily.dev/posts/unauthorized-group-has-gained-access-to-anthropic-s-exclusive-cyber-tool-mythos-report-claims-oow35kcqe) · TechCrunch · 30 upvotes · 17 comments --- Tags: [#security](https://daily.dev/tags/security), [#claude](https://daily.dev/tags/claude), [#anthropic](https://daily.dev/tags/anthropic), [#ai-safety](https://daily.dev/tags/ai-safety) [View this post on daily.dev](https://daily.dev/posts/anthropic-ran-133-million-contractor-chats-with-its-bioweapon-filters-off-ulyjzriyl) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Anthropic ran 133 million contractor chats with its bioweapon filters off","url":"https://daily.dev/posts/anthropic-ran-133-million-contractor-chats-with-its-bioweapon-filters-off-ulyjzriyl","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/anthropic-ran-133-million-contractor-chats-with-its-bioweapon-filters-off-ulyjzriyl"},"datePublished":"2026-08-15T14:25:17.478Z","dateModified":"2026-09-14T06:18:28.202Z","description":"Anthropic's second Risk Report reveals that bioweapon-blocking classifiers were disabled on its human feedback platforms for eleven months (May 2025 to April...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/dc0bcf82fd4faaf8ce28703e7d4d18fa?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/dc0bcf82fd4faaf8ce28703e7d4d18fa?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The Next Web","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The Next Web","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/tnw","url":"https://daily.dev/sources/tnw"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/anthropic-ran-133-million-contractor-chats-with-its-bioweapon-filters-off-ulyjzriyl","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"security,claude,anthropic,ai-safety","timeRequired":"PT7M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The Next Web","item":"https://daily.dev/sources/tnw"},{"@type":"ListItem","position":3,"name":"Anthropic ran 133 million contractor chats with its bioweapon filters off"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/anthropic-ran-133-million-contractor-chats-with-its-bioweapon-filters-off-ulyjzriyl#faq","mainEntity":[{"@type":"Question","name":"Why did Anthropic disable its bioweapon safety classifiers for contractors between May 2025 and April 2026?","acceptedAnswer":{"@type":"Answer","text":"An internal flag meant only for internal use accidentally switched off both the blocking classifiers and the logging mechanism on Anthropic's human feedback platforms, so flagged harmful content was never recorded or sent to review. Roughly 50,000 contractors generated about 133 million exchanges during this period, vetted by outside vendors that in many cases could not screen out even CB-1 level threat actors. Teams tracking AI vendor safety incidents can follow disclosures like this one on daily.dev."}},{"@type":"Question","name":"What did Anthropic find when it reviewed contractor conversations for bioweapon-related risk after discovering the classifier outage?","acceptedAnswer":{"@type":"Answer","text":"Claude Sonnet 5 scanned every human turn sent during the affected period and flagged 1,197 transcripts as high risk; 757 came from Anthropic's own teams and all but 62 of the remainder were deliberate red-teaming exercises. Staff manually reviewed the 62 plus 30 random red-team transcripts and found no clearly concerning misuse, though a handful of potentially dual-use conversations surfaced. Following how AI labs audit their own safety failures helps teams evaluating which providers to trust on daily.dev."}},{"@type":"Question","name":"What happened when contractors exploited a flaw to access Anthropic's Mythos Preview model outside their assigned work?","acceptedAnswer":{"@type":"Answer","text":"A few contractors at data-labelling vendors exploited a flaw in April 2026 to obtain an API key and used models, including the capable Mythos Preview model, outside their assigned scope. Mythos Preview sat exposed for roughly two weeks without blocking biological classifiers before Anthropic contained the access within 90 minutes of learning about it, with no model weights, customer data, or core networks compromised. daily.dev helps engineers weighing AI vendor risk stay current on incidents like this contractor access breach."}}]} ```