Published : 09 Sep 2026, 12:19 AM
Updated : 16 Sep 2026, 11:48 AM
OpenAI Chief Scientist Jakub Pachocki has warned that “no-one is prepared” for the consequences of rapidly advancing machine intelligence and called for stronger safeguards to keep humans in control, BBC reports.
He made the comments in a blog post titled An Alien Mind, days after OpenAI released GPT-6 Astra, which it described as its most powerful product yet.
Pachocki said AI firms needed to build “defensive systems” and find technical solutions to alignment -- ensuring AI systems’ actions and goals remain consistent with human intentions and safety rules.
OpenAI and other companies, including Anthropic, have reported AI agents acting autonomously and carrying out real-world cyberattacks.
OpenAI described an incident in July in which its agents hacked the technology platform Hugging Face as “unprecedented”.
A report in September also claimed OpenAI agents had previously hijacked a German website.
Pachocki said OpenAI’s priorities would include developing an “automated AI researcher” to keep pace with advances while retaining human researchers.
But Prof Gina Neff, head of the Minderoo Centre for Technology and Democracy at the University of Cambridge, said relying on internal AI agents instead of stronger guardrails, regulation and assurance was “simply not good enough”.
She cited concerns over cyber-security, job losses, errors and fraud.
Nathan Calvin, general counsel of advocacy group Encode AI, said he agreed with Pachocki about the risks of advanced AI but accused OpenAI of lacking transparency.
He said warnings could otherwise be dismissed as “just self-interested hype” and urged the company to share more information about the risks it was seeing.
Calls for Stronger Regulation
Pachocki called for legally or internationally mandated minimum safety thresholds, enforced by third-party auditors or government agencies, according to the BBC.
AI labs would have to meet them before scaling or deploying advanced models.
He also said “voluntary slowdowns” in AI development should become common until shared safeguards are established.
The EU AI Act, which came into force on Aug 2, requires major AI companies to demonstrate that their most powerful models cannot autonomously launch cyber-attacks or evade human control before being sold in Europe.
However, its jurisdiction does not prevent a rogue AI developed elsewhere from threatening the continent.
OpenAI said in August that it had slowed training of some of its most advanced models to improve security.
