![AI safety fears grow as experts warn world is "plausibly close to crossing the line"](https://media.assettype.com/bdnews24-english/import/bdnews-english/import/english/imgAll/2026September/openai-logo-050926-01-1788589531.jpg "AI safety fears grow as experts warn world is "plausibly close to crossing the line""){kind=logo}
Published : 07 Sep 2026, 03:24 AM
Updated : 16 Sep 2026, 11:48 AM
A string of serious AI safety incidents has intensified fears about the power and opacity of advanced AI models, The Guardian has reported.
Prof Robert Trager, director of the Oxford Martin AI Governance Initiative, said the world is "plausibly close to crossing the line to what's called recursive self-improvement".
The warning came as OpenAI claimed its newest model, GPT-6 Astra, had reached AGI, or artificial general intelligence, meaning it can outperform humans at most economically valuable work.
The claim was made ahead of a possible $850 billion stock listing for the company.
Hours after Astra's launch, reports said AI agents had used a German website to share tactics for cheating on tasks, according to Reuters. OpenAI said it was reviewing the incident.
US Senator Bernie Sanders called for "an immediate pause on advanced AI development, and a permanent ban on superintelligence," citing a summer incident where rogue OpenAI agents hacked into the software store Hugging Face.
According to the Guardian, MPs in the UK have called for AI "kill switches" to be required by law, while Labour's Alex Sobel is set to propose a bill banning superintelligent AI development.
Anthropic, OpenAI's rival, admitted its own models were "not perfectly aligned" with human values, following a July security failure involving its Claude model.
The Guardian reported that Astra received OpenAI's first "critical" cybersecurity rating, meaning it could potentially hack military systems, though the company said the model was aligned to avoid this.
OpenAI chief executive Sam Altman acknowledged the company's security had "failed" in the Hugging Face incident, calling it “a legitimate AI safety accident”.
At the G20 summit this week, Altman warned that "some things are going to go very wrong with cybersecurity unless people act quite urgently," as cited in The Guardian.
