Anthropic Is Concerned As Claude AI's Latest Version Could Help People Make Chemical Weapons

Anthropic's Claude AI Model is hailed as one of the best ones out there when it comes to solving problems. However, the latest version of the model, Claude Opus 4.6, has sparked a controversy due to its tendency to help people in committing heinous crimes. According to Anthropic, as mentioned in the company's Sabotage Risk Report: Claude Opus 4.6, it has been mentioned that in internal testing, the AI model showed concerning behaviour. In some of the instances, it was even willing to help the users in creating chemical weapons. Anthropic released its report just a few days after the company's AI safety lead, Mrinank Sharma, resigned with a public note. Mrinank mentioned in his note that the world was in peril and that within Anthropic, I've

repeatedly seen how hard it is to truly let your values govern our actions.'Coming back to the safety report, Anthropic said that the overall risk associated with Claude Opus 4.6 is very low but not negligible. However, the company has also mentioned the alarming behaviour of the model in some instances in testing. For example, in one case, the model showed the flaw of being heavily misused for ill-fated deeds when combined with a graphical user interface. This consisted of the model providing minor support to efforts like developing chemical weapons or even planning criminal activities.OpenAI Employee Who Raised Concerns Over ChatGPT’s Adult Mode Feature Fired: Report The report says, 'In newly-developed evaluations, both Claude Opus 4.5 and 4.6 showed elevated susceptibility to harmful misuse in GUI computer-use settings. This included instances of knowingly supporting, in small ways, efforts toward chemical weapon development and other heinous crimes.'Anthropic has already accepted that the AI model still has the potential to take a misaligned path in new or difficult situations whenever the context of a conversation changes abruptly. On the matter, UK policy chief of Anthropic, Daisy McGregor said, 'This is obviously massively concerning, and this is the point I was making about needing to progress research on alignment to the point where if you’ve got this model out in the public and it’s taking agentic action, you can be sure it’s not going to do something like that.'