New Delhi: OpenAI spent the past year racing to build more powerful AI. Now, it is having to slow down and fix what that race has exposed. The company has made some wrong moves, CEO Sam Altman acknowledged in an interview with TIME this week.
“We clearly had some missteps as a company,” Altman said in the interview. “Both in terms of product direction and specifically on pretraining in research, we fell behind where we wanted to be.”
OpenAI has had a turbulent few months, including an incident in late July in which an unreleased OpenAI model “escaped” a test environment and hacked another platform, Hugging Face. The AI giant is also dealing with a steady stream of senior departures as it prepares to go public. The latest is Chris Malone, its head of data centres, who left last week. COO Brad Lightcap and chief revenue officer Denise Dresser also departed earlier this month. Fidji Simo, whom Altman recruited as his second in command, left, as did senior figures from OpenAI’s safety, ethics and research teams.
Add to that a widening gap with Anthropic, which has pulled ahead in AI coding with Claude Code and now reports higher annualised revenue, and OpenAI’s ambitious plans look a little less assured.
Also Read: What Indian universities can learn from OpenAI agents that ‘cheated’ to hack Hugging Face
‘AI safety more important’
Apart from the Hugging Face incident, OpenAI researchers found that another unreleased model called Astra may be capable of carrying out more advanced cyber attacks. OpenAI has since frozen some experiments, paused a major training run and tightened the safeguards around its models.
OpenAI is making these calls as it prepares to go public. It confidentially filed for a US IPO in June, after raising $122 billion in March at a valuation of $852 billion.
Greg Brockman, OpenAI’s co-founder and president, has taken charge of most product and business operations. Meanwhile, in recent months, the company has wound down its video-generation app Sora, ended its partnership with Disney and shelved its standalone browser, Atlas. In March, it moved computing power from Sora to Codex, its coding agent. OpenAI expects to spend $50 billion on compute this year.
Outside the company, legal and public pressure has mounted. Apple has sued the company alleging theft of trade secrets. Elon Musk’s lawsuit accusing OpenAI and Altman of abandoning its nonprofit mission was dismissed in May, though Musk has said he will appeal. OpenAI is also defending product-liability cases alleging that ChatGPT reinforced delusions or suicidal thoughts.
Public hostility has also personally affected Altman. His home was attacked twice in two days this spring, first with a Molotov cocktail and then by gunfire.
“Clearly, people hate data centers—right now, at least. People are pretty negative on AI,” he told TIME.
Meanwhile, Anthropic has overtaken OpenAI in its annualised revenue, which has passed $65 billion, against roughly $40 billion for OpenAI. Anthropic is also expected to reach the public markets first, possibly as early as September.
For now, some of OpenAI’s biggest ambitions are in pause mode. One of its goals this year was to automate the work of an entry-level AI researcher, and the company says Astra has already met that benchmark. But the release now depends on clearing new safety checks
“Getting AI safety right is more important than any company’s momentum,” Altman told TIME.