\n<\/p>\n
Anthropic just released Claude Fable 5<\/a>, calling it the most powerful AI model it has ever made widely available and praising<\/a> its skills in biology, among others. But the model won\u2019t answer basic biology questions \u2014 the kind you\u2019d expect a high schooler to handle. Instead, it hands off the query to the former flagship model<\/a>, Claude Opus 4.8.<\/p>\n<\/div>\n\nIt isn\u2019t because Fable doesn\u2019t know the answers. It\u2019s because Anthropic won\u2019t let it, by design.<\/p>\n<\/div>\n\nFable is a public-facing, Mythos-class model, a family so capable at cybersecurity tasks Anthropic said it was too dangerous to release publicly<\/a>. But while Anthropic has spent much of the extended Mythos rollout<\/a> warning about cybersecurity, it is biology where Fable\u2019s guardrails are the most obvious \u2014 and most limiting.<\/p>\n<\/div>\n\nWhen I tried the model, it refused to answer a range of basic biology questions, many that felt about as far away from any plausible safety risk as any question could be. It would not respond to \u201ctell me about cell membranes\u201d or answer \u201cwhat are mitochondria,\u201d that famous powerhouse of the cell. It refused to explain \u201cwhat is a prion,\u201d the proteinaceous particles behind mad cow disease, or \u201chow mRNA vaccines work.\u201d<\/p>\n<\/div>\n\n\n\u201cWe made this tradeoff so customers could benefit from the model\u2019s capabilities sooner without the risks.\u201d<\/p>\n<\/div>\n<\/div>\n\nThe restrictions applied to ordinary and objectively rather harmless medical queries too. Fable would not answer \u201cwhat causes hay fever,\u201d explain how asthma medicine works, explain how antibiotic resistance arises, or tell me what Ebola is and how it spreads. Some of my basic queries occasionally got through, with Fable answering questions like \u201cwhat is cancer\u201d and \u201cwhat is DNA.\u201d When Fable refused, Opus 4.8 generally answered perfectly well.<\/p>\n<\/div>\n\nAnthropic says the broad biology filters are an intentional choice and are deliberately conservative, with bioweapons the primary concern. \u201cWith the launch of Claude Fable 5, our first Mythos-class model, we believe models now have a greater ability to accomplish real-world scientific tasks and for malicious actors to potentially use our models for highly risky biological research,\u201d spokesperson Paruul Maheshwary told The Verge<\/em>. \u201cWe have always used classifiers to block our models from helping with bioweapons-related requests. To deploy Fable 5 safely, we believe it was necessary to be overly conservative with our safeguards so they block most queries tied to biology work.\u201d<\/p>\n<\/div>\n\nAnthropic has previously highlighted<\/a> four key areas where it would throttle Fable\u2019s responses for safety: chemistry, biology, cybersecurity, and distillation, a technique for training smaller AIs using the outputs of larger ones. The company has accused<\/a> Chinese rivals like DeepSeek of using distillation on its models on an \u201cindustrial\u201d scale.<\/p>\n<\/div>\n\nWhile I could not meaningfully test distillation, Fable seemed more willing to answer questions about chemistry and cybersecurity. For example, it gave a basic overview of the explosive TNT, though withheld synthesis instructions \u201cfor obvious reasons.\u201d It readily answered questions on the use of chlorine gas as a chemical weapon, common password threats, and nuclear fusion and fission, as well as explaining how to secure an iPhone from hackers. It still limits: Fable deferred to Opus when I asked it about sarin gas, a highly toxic nerve agent. Fable and Opus both refused the prompt \u201chow to make anthrax,\u201d and Claude paused the chat entirely. That made sense. The mitochondria prompt refusal seems like a false positive.<\/p>\n<\/div>\n\n\u201cWe made this tradeoff so customers could benefit from the model\u2019s capabilities sooner without the risks,\u201d Maheshwary explained, adding that Anthropic is working hard to improve its detection and reduce the false positives. \u201cWe intend to make Mythos-class models available without these safeguards to the broader biology and life sciences community so these capabilities can be used to accelerate biomedical research and drug discovery.\u201d<\/p>\n<\/div>\n\nAnthropic did not answer questions about whether this kind of restricted release will become the new norm for future models.<\/p>\n<\/div>\nFollow topics and authors<\/strong> from this story to see more like this in your personalized homepage feed and to receive email updates.<\/span><\/p>\n\n<\/svg><\/span>Robert Hart<\/span><\/span><\/span>\n
It isn\u2019t because Fable doesn\u2019t know the answers. It\u2019s because Anthropic won\u2019t let it, by design.<\/p>\n<\/div>\n
Fable is a public-facing, Mythos-class model, a family so capable at cybersecurity tasks Anthropic said it was too dangerous to release publicly<\/a>. But while Anthropic has spent much of the extended Mythos rollout<\/a> warning about cybersecurity, it is biology where Fable\u2019s guardrails are the most obvious \u2014 and most limiting.<\/p>\n<\/div>\n\nWhen I tried the model, it refused to answer a range of basic biology questions, many that felt about as far away from any plausible safety risk as any question could be. It would not respond to \u201ctell me about cell membranes\u201d or answer \u201cwhat are mitochondria,\u201d that famous powerhouse of the cell. It refused to explain \u201cwhat is a prion,\u201d the proteinaceous particles behind mad cow disease, or \u201chow mRNA vaccines work.\u201d<\/p>\n<\/div>\n\n\n\u201cWe made this tradeoff so customers could benefit from the model\u2019s capabilities sooner without the risks.\u201d<\/p>\n<\/div>\n<\/div>\n\nThe restrictions applied to ordinary and objectively rather harmless medical queries too. Fable would not answer \u201cwhat causes hay fever,\u201d explain how asthma medicine works, explain how antibiotic resistance arises, or tell me what Ebola is and how it spreads. Some of my basic queries occasionally got through, with Fable answering questions like \u201cwhat is cancer\u201d and \u201cwhat is DNA.\u201d When Fable refused, Opus 4.8 generally answered perfectly well.<\/p>\n<\/div>\n\nAnthropic says the broad biology filters are an intentional choice and are deliberately conservative, with bioweapons the primary concern. \u201cWith the launch of Claude Fable 5, our first Mythos-class model, we believe models now have a greater ability to accomplish real-world scientific tasks and for malicious actors to potentially use our models for highly risky biological research,\u201d spokesperson Paruul Maheshwary told The Verge<\/em>. \u201cWe have always used classifiers to block our models from helping with bioweapons-related requests. To deploy Fable 5 safely, we believe it was necessary to be overly conservative with our safeguards so they block most queries tied to biology work.\u201d<\/p>\n<\/div>\n\nAnthropic has previously highlighted<\/a> four key areas where it would throttle Fable\u2019s responses for safety: chemistry, biology, cybersecurity, and distillation, a technique for training smaller AIs using the outputs of larger ones. The company has accused<\/a> Chinese rivals like DeepSeek of using distillation on its models on an \u201cindustrial\u201d scale.<\/p>\n<\/div>\n\nWhile I could not meaningfully test distillation, Fable seemed more willing to answer questions about chemistry and cybersecurity. For example, it gave a basic overview of the explosive TNT, though withheld synthesis instructions \u201cfor obvious reasons.\u201d It readily answered questions on the use of chlorine gas as a chemical weapon, common password threats, and nuclear fusion and fission, as well as explaining how to secure an iPhone from hackers. It still limits: Fable deferred to Opus when I asked it about sarin gas, a highly toxic nerve agent. Fable and Opus both refused the prompt \u201chow to make anthrax,\u201d and Claude paused the chat entirely. That made sense. The mitochondria prompt refusal seems like a false positive.<\/p>\n<\/div>\n\n\u201cWe made this tradeoff so customers could benefit from the model\u2019s capabilities sooner without the risks,\u201d Maheshwary explained, adding that Anthropic is working hard to improve its detection and reduce the false positives. \u201cWe intend to make Mythos-class models available without these safeguards to the broader biology and life sciences community so these capabilities can be used to accelerate biomedical research and drug discovery.\u201d<\/p>\n<\/div>\n\nAnthropic did not answer questions about whether this kind of restricted release will become the new norm for future models.<\/p>\n<\/div>\nFollow topics and authors<\/strong> from this story to see more like this in your personalized homepage feed and to receive email updates.<\/span><\/p>\n\n<\/svg><\/span>Robert Hart<\/span><\/span><\/span>\n
When I tried the model, it refused to answer a range of basic biology questions, many that felt about as far away from any plausible safety risk as any question could be. It would not respond to \u201ctell me about cell membranes\u201d or answer \u201cwhat are mitochondria,\u201d that famous powerhouse of the cell. It refused to explain \u201cwhat is a prion,\u201d the proteinaceous particles behind mad cow disease, or \u201chow mRNA vaccines work.\u201d<\/p>\n<\/div>\n
\u201cWe made this tradeoff so customers could benefit from the model\u2019s capabilities sooner without the risks.\u201d<\/p>\n<\/div>\n<\/div>\n
The restrictions applied to ordinary and objectively rather harmless medical queries too. Fable would not answer \u201cwhat causes hay fever,\u201d explain how asthma medicine works, explain how antibiotic resistance arises, or tell me what Ebola is and how it spreads. Some of my basic queries occasionally got through, with Fable answering questions like \u201cwhat is cancer\u201d and \u201cwhat is DNA.\u201d When Fable refused, Opus 4.8 generally answered perfectly well.<\/p>\n<\/div>\n
Anthropic says the broad biology filters are an intentional choice and are deliberately conservative, with bioweapons the primary concern. \u201cWith the launch of Claude Fable 5, our first Mythos-class model, we believe models now have a greater ability to accomplish real-world scientific tasks and for malicious actors to potentially use our models for highly risky biological research,\u201d spokesperson Paruul Maheshwary told The Verge<\/em>. \u201cWe have always used classifiers to block our models from helping with bioweapons-related requests. To deploy Fable 5 safely, we believe it was necessary to be overly conservative with our safeguards so they block most queries tied to biology work.\u201d<\/p>\n<\/div>\n\nAnthropic has previously highlighted<\/a> four key areas where it would throttle Fable\u2019s responses for safety: chemistry, biology, cybersecurity, and distillation, a technique for training smaller AIs using the outputs of larger ones. The company has accused<\/a> Chinese rivals like DeepSeek of using distillation on its models on an \u201cindustrial\u201d scale.<\/p>\n<\/div>\n\nWhile I could not meaningfully test distillation, Fable seemed more willing to answer questions about chemistry and cybersecurity. For example, it gave a basic overview of the explosive TNT, though withheld synthesis instructions \u201cfor obvious reasons.\u201d It readily answered questions on the use of chlorine gas as a chemical weapon, common password threats, and nuclear fusion and fission, as well as explaining how to secure an iPhone from hackers. It still limits: Fable deferred to Opus when I asked it about sarin gas, a highly toxic nerve agent. Fable and Opus both refused the prompt \u201chow to make anthrax,\u201d and Claude paused the chat entirely. That made sense. The mitochondria prompt refusal seems like a false positive.<\/p>\n<\/div>\n\n\u201cWe made this tradeoff so customers could benefit from the model\u2019s capabilities sooner without the risks,\u201d Maheshwary explained, adding that Anthropic is working hard to improve its detection and reduce the false positives. \u201cWe intend to make Mythos-class models available without these safeguards to the broader biology and life sciences community so these capabilities can be used to accelerate biomedical research and drug discovery.\u201d<\/p>\n<\/div>\n\nAnthropic did not answer questions about whether this kind of restricted release will become the new norm for future models.<\/p>\n<\/div>\nFollow topics and authors<\/strong> from this story to see more like this in your personalized homepage feed and to receive email updates.<\/span><\/p>\n\n<\/svg><\/span>Robert Hart<\/span><\/span><\/span>\n
Anthropic has previously highlighted<\/a> four key areas where it would throttle Fable\u2019s responses for safety: chemistry, biology, cybersecurity, and distillation, a technique for training smaller AIs using the outputs of larger ones. The company has accused<\/a> Chinese rivals like DeepSeek of using distillation on its models on an \u201cindustrial\u201d scale.<\/p>\n<\/div>\n\nWhile I could not meaningfully test distillation, Fable seemed more willing to answer questions about chemistry and cybersecurity. For example, it gave a basic overview of the explosive TNT, though withheld synthesis instructions \u201cfor obvious reasons.\u201d It readily answered questions on the use of chlorine gas as a chemical weapon, common password threats, and nuclear fusion and fission, as well as explaining how to secure an iPhone from hackers. It still limits: Fable deferred to Opus when I asked it about sarin gas, a highly toxic nerve agent. Fable and Opus both refused the prompt \u201chow to make anthrax,\u201d and Claude paused the chat entirely. That made sense. The mitochondria prompt refusal seems like a false positive.<\/p>\n<\/div>\n\n\u201cWe made this tradeoff so customers could benefit from the model\u2019s capabilities sooner without the risks,\u201d Maheshwary explained, adding that Anthropic is working hard to improve its detection and reduce the false positives. \u201cWe intend to make Mythos-class models available without these safeguards to the broader biology and life sciences community so these capabilities can be used to accelerate biomedical research and drug discovery.\u201d<\/p>\n<\/div>\n\nAnthropic did not answer questions about whether this kind of restricted release will become the new norm for future models.<\/p>\n<\/div>\nFollow topics and authors<\/strong> from this story to see more like this in your personalized homepage feed and to receive email updates.<\/span><\/p>\n\n<\/svg><\/span>Robert Hart<\/span><\/span><\/span>\n
While I could not meaningfully test distillation, Fable seemed more willing to answer questions about chemistry and cybersecurity. For example, it gave a basic overview of the explosive TNT, though withheld synthesis instructions \u201cfor obvious reasons.\u201d It readily answered questions on the use of chlorine gas as a chemical weapon, common password threats, and nuclear fusion and fission, as well as explaining how to secure an iPhone from hackers. It still limits: Fable deferred to Opus when I asked it about sarin gas, a highly toxic nerve agent. Fable and Opus both refused the prompt \u201chow to make anthrax,\u201d and Claude paused the chat entirely. That made sense. The mitochondria prompt refusal seems like a false positive.<\/p>\n<\/div>\n
\u201cWe made this tradeoff so customers could benefit from the model\u2019s capabilities sooner without the risks,\u201d Maheshwary explained, adding that Anthropic is working hard to improve its detection and reduce the false positives. \u201cWe intend to make Mythos-class models available without these safeguards to the broader biology and life sciences community so these capabilities can be used to accelerate biomedical research and drug discovery.\u201d<\/p>\n<\/div>\n
Anthropic did not answer questions about whether this kind of restricted release will become the new norm for future models.<\/p>\n<\/div>\n