Lisa Thiergart, CEO of the SL5 Task Force, says espionage against AI companies is no longer a hypothetical risk and that spies are already operating inside them.
AI security · Insider threat
Insiders, not just outsiders, are now a core security threat for frontier AI
SL5 Task Force CEO Lisa Thiergart argues that espionage against AI companies is no longer hypothetical and that protecting frontier models means planning for people who already have authorized access.
The shift in security focus
Traditional focus
Perimeter defense
Blocking external hackers and intrusions
→
Emerging focus
Insider risk
Staff with legitimate access who may work for hostile actors
What insiders can reach
Models
Frontier model access and weights
Infrastructure
Training systems and compute
Research
Internal research material and data
The case for the claim
Lab access controls and monitoring were largely built around external intrusion.
Insider risk, including authorized staff working for hostile actors, deserves core planning.
Frontier models are among the most valuable and strategically sensitive software assets today.
What remains unverified
The specific cases, companies involved and supporting evidence were not available.
Claims that spies are already inside AI companies cannot be independently confirmed.
Any conclusion about the scale of the problem is provisional.
What comes next
Labs face growing pressure to prove how they control insider access.
Limit and monitor
Internal access controls and oversight
Vet staff
Screening of people with sensitive access
Detect misuse
Spotting misuse of model weights and research data
Whether this becomes formal standards or policy depends on how much verifiable evidence of insider espionage emerges.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...