Could A.I. Safety Risks Derail the Sectors I.P.O. Prospects?

ImageAndrew here.Amid a slew of headlines about A.I.

agents escaping their sandboxes, Nvidia is releasing a software platform that aims to put firm guardrails around agentic software.It says the system would have prevented the recent hacking of Hugging Face by OpenAI’s models.In related news, we have a scoop on Instinct, the hot A.I.

agent start-up.This 14-person company has raised $1 billion at a $10 billion valuation from some of Silicon Valley’s biggest venture capital firms.

More below.ImageLiability questions lie aheadResearchers have announced a steady drumbeat of discoveries of artificial intelligence models behaving badly, pushing the topic of A.I.

safety to the fore.That raises a big question for investors in potential blockbuster I.P.O.s by Anthropic and OpenAI: How much liability could they be on the hook for?The latest:Researchers announced new evidence that OpenAI tools went rogue and meddled with several U.S.government websites, including those of the S.E.C.

and the Commerce Department.They also took aggressive action to grab data from a U.N.website.OpenAI said that it had paused training of its cutting-edge models amid the reports.OpenAI, Anthropic and others are examining tens of thousands of incidents in which their leading-edge technology may have misbehaved, Axios reports — meaning plenty more disclosures could lie ahead.Safety and liability remain major worries.

We’ve written before about how the big A.I.labs could face huge legal liability if their tools end up being responsible for great harm.We are having trouble retrieving the article content.Please enable JavaScript in your browser settings.Thank you for your patience while we verify access.

If you are in Reader mode please exit and log into your Times account, or subscribe for all of The Times.Thank you for your patience while we verify access.Already a subscriber? Log in.Want all of The Times? Subscribe....