OpenAI warns AI-enabled cyber attacks could become persistent

OpenAI’s chief global affairs officer, Chris Lehane, has warned that organisations should prepare for increasingly persistent cyber attacks powered by advanced AI. He said improving model capabilities are changing the threat landscape, particularly as powerful models become capable of planning and executing more complex offensive activity.

Lehane also called for mandatory safety standards for frontier AI models before release. The comments come as OpenAI strengthens safeguards around models showing advanced cybersecurity capabilities.

Key facts

• OpenAI says increasingly capable AI could enable sustained and persistent cyber attacks.
• Lehane highlighted particular concern about advanced models becoming widely accessible.
• OpenAI has slowed development of some frontier models while additional safeguards are implemented.
• OpenAI is strengthening monitoring, alignment and containment controls.

Our take

This is relevant well beyond AI developers. If offensive AI capabilities improve, organisations may face attackers that can identify vulnerabilities, adapt techniques and continue probing systems with much less human involvement.

For New Zealand organisations, defensive automation will become increasingly important, but so will basic controls. Strong identity management, patching, segmentation, logging and tested incident response remain essential because AI gives attackers more ways to exploit weaknesses rather than making those weaknesses disappear.

Sources

• The Guardian: OpenAI leader warns of threat of persistent AI cyber attacks
• OpenAI: Pacing model development in an era of cyber-critical capabilities