AI companies warn of existential risks from their own technologies

5 articles · Updated · BBC.com · Apr 29

Anthropic claims its new model, Claude Mythos, can find cybersecurity bugs surpassing human experts and has partnered with over 40 organizations to patch vulnerabilities.
Critics argue these apocalyptic warnings exaggerate AI's power to distract from current harms, boost stock prices, and influence regulation, despite doubts from security experts about the companies' claims.
Industry leaders, including OpenAI and Anthropic, have a history of such warnings, fueling narratives that only they can manage AI risks, while broader concerns like environmental impact and mental health remain under-addressed.

Are AI companies exaggerating risks to boost their power and profits, or are the threats as catastrophic as they claim?

How credible are Anthropic's claims about Claude Mythos' cybersecurity abilities, given the lack of standard metrics and external skepticism?

Will the current regulatory patchwork—federal delays, state action, and industry lobbying—ever ensure real AI accountability and safety?

What role should independent oversight play as AI companies abandon safety guarantees in the race for market dominance?

Why are immediate harms like environmental destruction and job loss overshadowed by apocalyptic AI scenarios in public discourse?

With AI now central to military operations and linked to civilian casualties, who is accountable for errors and unintended harm?