The AI control gap: who gets to say an AI system is safe?
A SiliconANGLE analysis published October 10 argues there is a "control gap" between what AI can do and what the evidence supports trusting it to do, noting that a good demo does not prove a system will stay within the limits of an intended deployment. Drawing on conversations with Appian CEO Matt Calkins, it calls for qualified independent experts to shape public standards, expert assessors to access and challenge vendors' evidence, and an accountable public authority able to require remediation, restrict or stop high-risk activity. It notes that the September 29 White House accord requires an independent committee of each company's board to provide external assessment and oversight, but creates no public authority with mandatory access, compliance conditions and penalties; it also recounts an All In podcast exchange in which David Sacks said external audits should be done by professional auditors rather than NGOs and Chamath Palihapitiya hinted EY (also Anthropic's auditor) would announce something soon, and cites Qualitate customer data on worries about undocumented "agent sprawl."
If safety evidence is mainly self-attested by vendors, authorizing high-risk AI deployments lacks an enforceable accountability step; this debate shapes release conditions and enterprise compliance costs.