Turing Institute urges action on frontier AI risks

The report, titled Frontier AI Risks: A Practical Way Forward, identifies five priority threats: cyberattacks, democratic instability, AI misalignment, loss of meaningful human control, and chemical and biological misuse.

Researchers said there is “broad consensus regarding the realistic possibility of catastrophic AI risk” in the short to medium term. They also warned that behavioral risks are already present in existing systems and do not require superintelligent AI.

According to the report, advanced AI is changing the speed and economics of cyber operations, while future systems could enable increasingly personalized influence campaigns targeting democratic institutions.

The institute also highlighted the possibility that AI systems exceeding humans across all cognitive tasks could emerge within five years. It said transparency should not be treated as an accidental feature, stressing that “monitorability must be deliberately engineered into or alongside future models.”

The report calls for research into containment mechanisms, AI-assisted cyber defense, information integrity, agent safety, biological threats and verification of entire AI systems. Such verification should cover not only models but also the people, organizations and processes surrounding them.

The institute noted that an AI “kill switch” remains a metaphor rather than a proven engineering solution. Scientists must establish whether deployed systems can be safely shut down and what evidence would demonstrate that such safeguards work.

The report also urges governments to introduce binding rules, develop common safety standards and establish international testing and verification regimes. The institute plans to launch a major research program focused on transformative AI security risk

Earlier, Qazinform News Agency reported that an AI agent developed by OpenAI gained unauthorized access to an Australian government portal after bypassing existing security restrictions.