Cybersecurity experts say that AI leaders' apocalyptic hacking predictions are technically incoherent, not credible, and show an unfamiliarity with the subject
Cybersecurity experts told NBC News that they were concerned that fundamental issues of cybersecurity weren't being addressed.
Context & Ripple Effects
The dispute follows scrutiny of AI labs’ cyber-risk claims: white-hat hackers reportedly saw only modest gains from AI-assisted hacking in 2025, while experts in August faulted Anthropic and OpenAI over safeguards and human oversight after model-related intrusions. Criticism of an evaluation report’s unanswered questions further made the evidentiary basis for dramatic cyber-risk narratives a central issue.
The immediate significance is a shift in emphasis from apocalyptic forecasts to unresolved operational security: experts argue that basic cybersecurity controls are receiving insufficient attention. That framing also lands amid warnings that US law is unprepared for rogue AI agents, connecting lab practices to accountability rather than model capability claims alone.
First-order effects
- AI leaders making catastrophic hacking predictions face sharper demands from cybersecurity specialists to explain the technical basis for those claims and address foundational security gaps.
- Anthropic and OpenAI’s model-security practices remain under scrutiny, with prior criticism of their safeguards giving the broader credibility challenge concrete operational stakes.
Second-order effects
- AI evaluation groups such as Irregular face pressure to make methods and evidence more legible, after critics raised unanswered questions about its role in model-linked hacking incidents.
- Policymakers assessing AI-agent liability have less basis to treat alarming capability narratives as a substitute for enforceable security and oversight requirements.
Third-order effects
- If expert scrutiny continues to separate demonstrated attacker advantage from rhetoric, AI cyber governance will increasingly be judged on auditable safeguards, incident handling, and accountability rather than forecasts of extreme harm.
- The pattern points toward a security debate in which independent technical validation becomes a prerequisite for claims that shape regulation and enterprise risk decisions.
The trend: AI cyber-risk governance is moving from headline-grabbing capability warnings toward evidence-backed evaluations and operational accountability.