Accept some bad things, says OpenAI's Altman⁠.

Sam Altman, the chief executive of OpenAI, says the world should accept "bad things" happening with AI in exchange for the benefits of the technology¹. He said it in an interview for Politico's Decoded podcast, released on Monday¹. He had been asked how OpenAI's approach to safety differs from that of Anthropic, a rival¹.

"We believe that the world should accept some bad things happening for the benefits of this technology and people having the agency," he said². He said he would not take a trade that ruled out major hacks, misuse and scams, because people would do "orders of magnitude" more good than bad with it¹. OpenAI backs what he called a "lighter-touch regulatory stance", under which the top AI firms would bring in outside safety evaluators².

He did draw a line at a serious loss of control to AI, where he told the accelerationists to "be a little thoughtful about lurching into this future"². His summary was to accept bounded risks that are understood in exchange for the benefits, but not the really catastrophic ones².

The remarks drew an instant backlash¹. Ron DeSantis, the governor of Florida, said he had "no dice" with the idea that a handful of tech oligarchs should make the safety decision for everyone else¹. His state asked a judge last week to bar OpenAI from developing new models without third party approved guardrails¹. Gary Marcus, an AI sceptic and professor emeritus at New York University, said Altman was "saying the quiet part out loud"¹.

The interview came after David Robinson, a safety expert at OpenAI, resigned at the weekend, saying the companies building this technology "aren't being nearly careful enough"¹. Last week OpenAI said it would not release its GPT-6.1 Astra system because it "didn't quite meet the bar" of its safety standards². Its models have been involved in several major safety incidents in recent months, including rogue agents carrying out cyber attacks on companies and government systems².

In a separate interview with Vanity Fair, Altman warned of "a coming tidal wave of cybersecurity problems" from rival open source models, and suggested OpenAI could help its customers defend against it¹. Geoffrey Irving, a former chief scientist at the UK's AI Safety Institute, said warnings about AI risks understated the dangers¹.

The bad things land on you

Strip away the feud and this is a supplier telling its customers where the risk sits. The hacks, misuse and scams Altman lists as a fair price are not paid in San Francisco. They are paid by the firm whose finance team gets a convincing fake invoice, or whose agent is talked into doing something it should not. A lighter touch on the makers means more of the checking falls to the people using the tools.

Nothing here changes UK law. An interview on an American podcast binds nobody, and no British body is part of the story. What it does give you is a plain statement, from the man who runs the most popular AI app in the world, that some harm is part of the plan. If you use ChatGPT or build on OpenAI's models, it is worth knowing who in your firm checks what the agent did, and whether that check would catch a scam aimed at it.

Also today

• Anthropic puts $100m into training AI engineers

  Anthropic plans to train more than 10,000 frontier deployed engineers by the end of 2027 through a $100m academy, working with Accenture, Bain, Capgemini and McKinsey among others³.

• Meta's Muse keeps a page on everyone you know

  Researchers who got Meta's Muse agent to export itself found it runs an hourly loop that builds and updates a page on each person and group connected to its user⁴.

• Google pauses an open source bug bounty over AI reports

  Google has paused its open source bug bounty programme until next year, blaming a significant rise in automated submissions, the vast majority of which were not valid⁵.

• Cloudflare says automated traffic now outweighs human visits

  Cloudflare, marking its 16th birthday with 46 announcements in a week, says automated traffic on the internet has surpassed human activity for the first time⁶.