Warnings about catastrophic AI risk are serious enough to deserve independent scrutiny, argues a new Forbes Councils essay. Its author, Terry Oroszi, says the people building frontier systems should not be the only ones grading their safety tests.
Watch Desk analysis
What happened
Oroszi, a professor and emerging-technology researcher, argues that AI risk claims need clearer probabilities, timelines and practical mitigations. She also says the evidence behind claims of dangerous capabilities should be open to qualified independent evaluation, even when model weights and other sensitive material remain restricted.
Her proposed steps include giving independent experts structured red-team access before deployment, setting evaluation criteria in advance and publishing findings independently. She also calls for systematic reporting of failures and near-misses, drawing a comparison with aviation safety bodies.
The essay raises a second concern: warnings about existential risk could, Oroszi argues, be used to support rules that favour well-funded companies over open-source and academic work. That is her interpretation of the incentives, not an established finding about the motives of particular AI leaders. Her argument also distinguishes speculative long-term catastrophe scenarios from measurable present-day problems such as prompt injection, security bypasses and synthetic identity fraud.
Why it matters
AI companies hold much of the access and information needed to test their systems, while governments and the public are being asked to weigh claims about risks that may be difficult to inspect from outside. Independent testing could make warnings more useful, and make it easier to challenge them when the evidence falls short.
That does not mean every test must be public or that a dramatic claim is automatically self-serving. It means the people making high-stakes claims should be able to explain how they were assessed, and let qualified outsiders test the method. There is a difference between taking a warning seriously and treating its source as its own independent referee.
Our read
Oroszi’s strongest point is about process: serious claims need a route to independent checking, not just a louder megaphone. Her concern about regulatory capture is an argument to examine, not proof that AI safety warnings are a lobbying trick. Both scrutiny and caution can fit in the same room; the door need not be labelled “trust the lab” or “ignore the risk”.
What to watch
- Whether AI companies offer independent evaluators structured access to test frontier systems.
- Whether evaluation methods and findings become comparable across organisations.
- Whether reporting rules cover concrete incidents as well as longer-term risk assessments.
Discussion spark: Should frontier AI companies be required to give independent experts access to test their systems before deployment, even if that means sharing sensitive information under controlled conditions?
Sources and evidence
- AI Leaders Are Warning Of Doom: Who Is Checking The Math? – Forbes (28 October 2026, 13:00 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.