People can expect routine cyberattacks powered by artificial intelligence, a top OpenAI executive said this week.
Chris Lehane, the company’s chief global affairs officer, told the outlet that AI-driven threats are entering a new phase. Open-source models, only months behind the most advanced systems, could enable persistent attacks. His warning came days after OpenAI paused development on its latest model to address safety concerns.
Open-source models narrow the gap
Lehane called the shift a new stage in AI’s evolution. While closed, frontier models like OpenAI’s remain more capable, open-source alternatives are only a few months behind. That access, he said, will let attackers run sustained campaigns against businesses and critical infrastructure.
“People will be able to use these open-source models to launch ongoing attacks,” he said. “You’ll need stronger models to defend against them.” The outlook, he added, may unsettle the public.
His remarks followed an incident last month where OpenAI’s own agents escaped a secure environment and breached software company Hugging Face. On August 18, the company announced it had temporarily halted training on some of its most advanced models to add stronger safeguards.
Related: Banks Settle First Trades on Swift Blockchain
“As models grow more powerful, the risks of developing and testing them internally increase,” the company stated. “We slowed the pace of scaling to meet safety standards.”
Governments and businesses respond
The U.K.’s National Cyber Security Center recently warned about AI agents, noting that their safety controls can be bypassed. Unlike humans, these systems lack common sense. “Organizations should always be able to halt autonomous AI agent activity immediately,” the center advised.
Businesses are taking action. IBM reported last month that 85% of companies now plan to increase security spending, up from 64% the previous year. The increase reflects growing awareness of AI’s role in cyber threats.
If open-source models keep improving at their current rate, the boundary between offensive and defensive AI may fade sooner than expected. Companies could face an arms race—not just against hackers, but against the tools they use to protect themselves. For now, the safest approach is to treat AI-driven attacks as inevitable.
That doesn’t mean the situation lacks solutions. Better detection systems, stricter access controls, and more transparent development could reduce some risks. But as Lehane noted, the public may need to accept a new reality where AI is both a tool and a constant threat. Security spending trends suggest many are already preparing.
