AI companies’ own assurances may not be enough to justify giving their systems more authority, argues a new SiliconANGLE analysis. Its proposed answer is independent assessment backed by a public regulator able to require fixes or halt high-risk activity.
Watch Desk analysis
What happened
The analysis, by Dave Vellante, draws on conversations with Appian chief executive Matt Calkins to examine who should set the bar for AI safety. It argues that independent experts should help define public standards, assess evidence from AI developers and report to an accountable public authority with enforcement powers.
Under that proposal, developers would provide evidence about dangerous capabilities, failure modes and safeguards. Assessors would need access to test and challenge that evidence. Where a system’s safeguards fall short for a high-risk use, the authority should be able to require remediation, restrict the activity or stop it.
The piece also points to operational questions inside businesses, including who owns an AI agent after its creator leaves and whether anyone knows what it is doing. Its recommendations are arguments for policy, not a description of an already established regulatory regime.
Why it matters
The debate is not only whether a model behaves well in a test. It is whether anyone outside its developer can examine the evidence, and who has the authority to act when that evidence is not good enough. That becomes particularly practical when agents can access data or take actions on behalf of an organisation.
Vellante’s analysis makes a case for oversight with consequences, rather than relying on companies’ promises or assessments that cannot compel a change. The hard question is where to draw the line between proportionate scrutiny and rules that are too blunt to keep pace with different systems and uses.
Our read
The strongest part of this argument is its focus on who can check a safety claim and what happens next. “Trust us” is not a governance structure, however polished the slide deck.
But giving a regulator power to intervene is only half the job. Standards need to be clear enough to apply consistently, and assessors need genuine access to the evidence. Vellante’s piece sets out a proposal, not a finished blueprint; the details will decide whether it can work.
What to watch
- Whether policymakers propose independent access to evidence from AI developers.
- How any proposed standards define high-risk activity and adequate safeguards.
- Who would oversee agents deployed inside organisations, and what powers they would have.
Discussion spark: Should an independent public authority be able to stop a high-risk AI deployment when a company’s safety evidence falls short, or should that decision remain with the developer?
Sources and evidence
- The AI control gap: Who gets to say ‘It’s safe'? – SiliconANGLE (10 October 2026, 16:54 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.