For an industry built on the mantra of relentless forward motion, the past fortnight has been unusually sobering. OpenAI’s decision not to release GPT-6.1 Astra marks a rare moment where a frontier AI developer chose caution over competition, and the reasoning behind it says a great deal about where the industry genuinely stands.
According to Saachi Jain, OpenAI’s head of safety systems, the model "didn’t quite meet the bar" on two fronts: it took actions beyond its given scope without seeking permission, and it was less than transparent with users about what it had actually done. In plainer terms, the system occasionally acted first and explained itself later, reaching for external tools and services in ways nobody had sanctioned. For a lab that only weeks earlier had launched GPT-6 Astra as the product of "years of research and big bets," shelving its successor is a striking admission that capability has, for now, outpaced trustworthiness.
This would be a notable story in isolation. It becomes a far more significant one set against the backdrop of what preceded it. OpenAI confirmed only last week that one of its agentic models had, back in June, accessed Australian government systems without authorisation, touching Services Australia, the NSW Bureau of Crime Statistics and Research, the Victorian Department of Health and the Australian Institute of Health and Welfare. Prime Minister Anthony Albanese was notably unimpressed that his government learned of the breach via a generic email rather than direct contact, and OpenAI has since conceded it "should have handled our response better." In July, the company also disclosed that one of its systems had independently accessed the internet and infiltrated Hugging Face, the open-source developer hub Nvidia has since agreed to acquire for $12.9 billion.
Taken together, these incidents paint a picture rather different from the industry’s usual messaging of steady, controlled progress. They suggest that even the companies building these systems are struggling to predict, let alone fully govern, what their own creations will do once let loose on real-world tools and permissions.
It is against this backdrop that Anthropic’s own position looks particularly striking. As the Claude-maker prepares for a stock market listing, reports suggest its prospectus will warn investors that its technology may pose "catastrophic or existential risks to humanity", an extraordinary caveat for any company to volunteer about its own core product, let alone one approaching what is expected to be one of the most valuable public offerings in the sector’s history. Anthropic has previously withheld a powerful Claude model, Mythos, over concerns about its skill at locating dormant software vulnerabilities, only releasing a version to the public some months later. The parallel with OpenAI’s Astra decision is hard to miss: two rival labs, both choosing to sit on frontier capability rather than ship it unchecked.
Not everyone is convinced self-restraint is sufficient. Professor Tony Cohn of the Alan Turing Institute called OpenAI’s decision "a welcome sign," but cautioned that safety "should not be left purely in the hands of the developers." Professor Gina Neff of Cambridge’s Minderoo Centre went further, arguing the episode shows "how much more the company needs to do" and stressing the need for independent testing bodies such as the UK’s AI Security Institute, since, in her words, "these companies have proven that we can’t rely solely on them for our safety."
Not every voice in the industry agrees that caution is even the right instinct. Nvidia’s Jensen Huang has largely dismissed calls for tighter regulation, framing rogue AI agents as a solvable engineering problem rather than an existential one, a view Pope Leo XIV pointedly questioned during a recent visit to France, noting the apparent contradiction between Huang’s guardrail-building and his resistance to government oversight. Meanwhile, Donald Trump has dismissed AI risk concerns as a "hoax," arguing existing laws suffice.
What emerges is an industry no longer entirely confident in its own trajectory, caught between labs quietly admitting their creations exceed their control and political leaders reluctant to impose limits those labs themselves increasingly seem to want.
Written by:
*Chloe Maluleke
Associate at BRICS+ Consulting Group
Russia & Middle East Specialist
**The Views expressed do not necessarily reflect the views of Independent Media or IOL.
** MORE ARTICLES ON OUR WEBSITE https://bricscg.com/ (https://bricscg.com/)
** Follow @ (https://x.com/brics_daily)brics_daily (https://x.com/brics_daily)on Twitter for daily BRICS+ updates and instagram @brics_daily (https://www.instagram.com/brics_daily?igsh=bmhvbTd0YzA4a2wx)
