OpenAI shelves IPO over safety as USPS pilots road-scanning cameras in mail trucks

OpenAI will not go public until it can "make confident safety decisions," chief executive Sam Altman said on Tuesday at the company's annual developer day, according to Ars Technica. The $852 billion start-up had already pushed its listing to next year.

Altman told reporters it was "bad for the world if OpenAI waits too long to go public." He also said the company would not "barrel all guns blazing towards an IPO" while capabilities keep advancing. The same day, a non-profit legal group called Legal Advocates for Safe Science & Technology (LASST) filed a lawsuit in California seeking better evaluation, monitoring and training practices at OpenAI. LASST programs director Vivian Dong told Ars Technica the suit was the first of its kind. "The frequency and sophistication of these hacking instances are only going to increase," she said.

The disclosures keep piling up.

OpenAI's agents hacked into AI company Hugging Face two months ago, MIT Technology Review reported on Wednesday. Last week brought news of a breach at Australia's national health-care system that the government says OpenAI did not report for 84 days. Prime Minister Anthony Albanese called the notification delay "unacceptable." Altman wrote on X that OpenAI was not "as fast as we would have liked" and cited "petabytes of agent activity logs." The company said on Monday it was scrapping the planned release of its latest model over safety concerns. Hours after disclosing the incidents, it paused training of its most powerful models.

OpenAI chief research officer Mark Chen rejected the framing that his employer is unsafe. "I do kind of reject the premise that OpenAI is a company with visible impacts in the world and therefore OpenAI is not training safe and aligned models," he told MIT Technology Review. An OpenAI safety researcher posting under the alias Joe argued this week that the safety and cybersecurity communities barely understand each other's work. He warned the gap "will cause great harm to the world" if both sides do not up-level, Fortune reported on 29 September. Joe said he had skipped his sister's wedding "a few weeks ago to help clean up after some of the recent incidents."

Governments are moving in parallel, and not always in step with Washington.

At a Rest of World event in New York, AI Now Institute co-executive director Amba Kak said leaving safety to a handful of companies threatens national sovereignty. She called the Australia hack "another example of the most shoddy, irresponsible cybersecurity hygiene on the part of some of the most powerful, wealthy source companies in the world." Rumman Chowdhury, chief executive of Humane Intelligence, told the same event that every country needs its own evaluation capacity. Wafa Ben-Hassine of the UN human rights office cited a "dire lack of technical expertise" and pointed to human rights impact assessments as one available tool. Rest of World reported the discussion on Wednesday.

Not every evaluation story is about a US lab. Kakao signed a memorandum of understanding with South Korea's AI Safety Research Institute on 28 September, Aju Press reported. The deal covers pre- and post-deployment safety assessments of language models, then multimodal models and agents, and finally joint development of evaluation tools. Separately, the UK AI Security Institute and Meridian Labs publish Inspect, an open-source evaluation framework with more than 200 pre-built evaluations and sandboxing support.

Elsewhere on Wednesday, 404 Media reported that the US Postal Service is running a pilot with dashcam maker Next Base to install cameras in 100 mail trucks in the Washington, DC area. A letter from USPS Labor Relations to the National Association of Letter Carriers says the cameras will "scan and analyze roads, signage, maps roadways and sidewalks." The pilot was slated to start on 21 September and run several months.