Anthropic CEO Dario Amodei: 25% Chance AI could be catastrophic, 75% chance it could thrive

Anthropic CEO Dario Amodei has highlighted both the risks and potential benefits of artificial intelligence, stating there is a 25 per cent chance that AI could lead to catastrophic outcomes, but a 75 per cent chance it could produce highly positive results, at the Axios AI + DC Summit.

Amodei explained that by “really, really badly,” he was not referring to minor issues like autocorrect errors, but to scenarios capable of destabilising societal systems, creating existential threats, or leading to severe misuse of AI.

He has previously warned that AI could replace up to half of all entry-level white-collar jobs and expressed concern over the export of advanced chips to China, signalling the economic and geopolitical stakes of AI advancement.

By framing the discussion probabilistically, Amodei emphasised that while risks exist, human choices and governance can influence outcomes, offering a balanced perspective that recognises danger but also potential opportunity.

Meanwhile, it was reported earlier this month that Anthropic PBC has agreed to a minimum payout of $1.5 billion, plus interest, to resolve a class action lawsuit brought by authors who accused the AI start-up of illegally scraping millions of copyrighted books to train its language models.

Among the largest AI-related settlements

Filed on Friday in a San Francisco federal court, the settlement ranks as one of the largest agreements to date involving AI and intellectual property. A trial had been scheduled for December, with plaintiffs seeking damages that could have reached trillions of dollars, potentially jeopardising the company’s future.

Compensation for millions of writers

The settlement covers claims from up to seven million authors, with initial payouts expected at around $3,000 per book for an initial set of 500,000 works. If additional claims are approved, the total compensation could increase further. Anthropic has also pledged to delete any disputed data it was alleged to have obtained unlawfully.

Legal representatives for the authors described the settlement as unprecedented. Justin Nelson of Susman Godfrey said, “This deal sends a strong message to AI companies and content creators that using copyrighted material from pirate sources is unacceptable.”

While the financial impact is substantial, legal analysts noted that Anthropic may have avoided a far worse outcome. Tod Cohen, partner at Manatt, Phelps & Phillips, commented, “The settlement represents less than one per cent of the company’s valuation. It appears to be a favourable result for Anthropic, as it avoids dismantling its models and only needs to remove the contested data.”