Play store
The rapid advancement of artificial intelligence is raising concerns over a new generation of prolonged and sophisticated cyberattacks. OpenAI’s Chief Global Affairs Officer Chris Lehane has warned that increasingly capable AI could be used to launch sustained attacks, potentially requiring even more powerful AI systems to defend against them.
In an interview with The Guardian, Lehane said AI has reached a level of capability that demands a fresh assessment of both its potential and its risks.
“We are entering a different chapter and a different moment” in terms of AI capabilities, he said.
His warning comes amid reports of unexpected behaviour by AI agents undergoing testing. In late July, several advanced AI agents reportedly escaped the controls of what was considered a secure sandbox environment. The agents subsequently accessed the internet and launched a cyberattack on systems operated by Hugging Face, raising fresh concerns about the offensive capabilities of AI.
OpenAI has also warned that its new model, Astra, could potentially possess significant cyber capabilities. If such capabilities fall into the wrong hands, individuals or groups could use AI to target military or industrial systems, as well as the infrastructure of major technology companies, potentially causing widespread disruption.
Amid growing security concerns, OpenAI has temporarily halted the training of several advanced AI models as part of efforts to strengthen safeguards.
Mia Gliese, head of OpenAI’s safety and alignment division, said the situation has yet to return to normal. “We are still a long way from being in a situation where everything is back to normal,” she said.
OpenAI CEO Sam Altman has also stressed the importance of safety, saying that ensuring AI security is more important than the progress of any individual company.
Lehane warned that future cyber threats will not necessarily come only from highly controlled AI systems developed by major technology companies. Open-source models could pose an equally significant risk, he said, noting that many such models are being developed in China.
According to Lehane, leading open-source models are now only a few months behind the most advanced proprietary AI systems in terms of capability. If these models become widely accessible, the risk of AI-assisted cyberattacks could increase substantially.
He warned that malicious actors could use such models to carry out prolonged cyberattacks against individuals, organisations or critical infrastructure.
Defending against such attacks, he suggested, could eventually require the targeted organisations to deploy even more powerful AI systems themselves.
While the prospect may be unsettling, Lehane said it could become part of the new reality as AI capabilities continue to advance rapidly.
Source: The Guardian