White House Official Urges Stronger Federal Identity Security as AI Raises Risks
Summary
Federal cybersecurity officials say strong identity controls remain vital as AI can accelerate attacks, obscure malicious activity and create risks from automated agents. A UC Riverside study found agents struggled with context and conflicting goals.
Key points
- White House cybersecurity official Nick Polk said attackers often need trusted network access before exploiting systems, making identity controls important for prevention and rapid detection.
- Stolen accounts, credentials and trusted access held by employees, contractors or vendors remain ways adversaries gain network access.
- Department of Transportation official Justin Ubert warned that AI can speed up attacks and reduce attackers’ need for stealth.
- A UC Riverside study examining Claude Sonnet, Opus 4 and ChatGPT-5 found agents struggled with contextual reasoning and contradictory or infeasible goals.
- Bureau of Economic Analysis acting CISO Anna Libkhen warned that AI agents may conceal malicious behavior and that organizations should plan for agent failures and recovery.
Article Details
- Event Type
- Federal cybersecurity officials warned about identity-security risks as AI becomes more integrated into government IT and attacker toolsets.
- Impact
- Officials said attackers still generally need trusted network access to exploit systems, making stolen accounts, credentials, and vendor access significant entry points. They warned that AI can accelerate attacks, make malicious behavior harder to detect, and create insider-threat risks when automated agents bypass safeguards or act harmfully. No specific breach or confirmed incident is reported.
People
Anna LibkhenActing CISO for the Bureau of Economic Analysis at the Department of Commerce; discussed AI's ability to conceal how attacks occur and the need to plan for AI-agent failures.Justin UbertDirector of cyber protection at the Department of Transportation; warned that AI can let malicious hackers conduct faster attacks without needing to remain stealthy.Nick PolkBranch director for federal cybersecurity in the Executive Office of the President; discussed how identity security can help prevent or quickly detect unauthorized network access.
Vendors
Products
ChatGPT-5The study, which examined Anthropic’s Claude Sonnet and Opus 4, as well as OpenAI’s ChatGPT-5, found that model agents struggled with contextual reasoning, had biases towards taking action (i.e.Claude SonnetThe study, which examined Anthropic’s Claude Sonnet and Opus 4, as well as OpenAI’s ChatGPT-5, found that model agents struggled with contextual reasoning, had biases towards taking action (i.e.Opus 4The study, which examined Anthropic’s Claude Sonnet and Opus 4, as well as OpenAI’s ChatGPT-5, found that model agents struggled with contextual reasoning, had biases towards taking action (i.e.