Tech giant Anthropic said it banned several accounts for potentially misusing their AI models to help develop biological weapons. It comes after a former employee warned AI could kill humanity in a matter of years. Matt Rivers reports.

@MattRiversABC According to Anthropic, the group was developing a guided rocket, a ballistic missile intended to reach more than 2,000 km, and a family of missile concepts that included a hypersonic glide vehicle variant.

@MattRiversABC Script is so ridiculous & obvious anymore. They create fear to trick you in support policies circumvent The Constitution not protect you. No coincidence not 1 arrest from any policy the gov't literally caught doing is also the backdoor or technicality used in all of them....

@MattRiversABC Just as Attorneys General ordered OpenAI to retain logs, they should target Anthropic. Spreading bioweapon claims without logs borders on market manipulation. Framing threats without data won't work: finance doesn't buy hype narratives—it buys contracts and audited data. #AI

@MattRiversABC Where have heard this bullshit before oh right American mainstream media and politicians and experts on tv now on x about wmd in Iraq and Iran .

@carlquintanilla @MattRiversABC But think about the shareholder value!

@MattRiversABC Sounds about right. But hey, once again, thank goodness for science and “the experts” for thinking they could play God

@MattRiversABC WOW !

@MattRiversABC AI can transform humanity but it must never become a tool for catastrophic harm. Anthropic says it blocked attempts involving potentially dangerous biological research, underscoring why strong safeguards, transparency, and responsible development are essential.

@MattRiversABC That is the Anthropic bomb to scare people

@MattRiversABC Organizations should establish clear, transparent protocols for auditing model misuse. What specific indicators are currently used to identify attempts to develop biological weapons?

@MattRiversABC The intersection of AI and biological security is a critical area of concern right now. We actually went deeper on this here: