Current and former OpenAI and Google DeepMind researchers are warning that AI labs are not doing enough to prepare for self-improving systems. Their accounts, gathered for a new public-awareness project, put a pointed question to the field: who gets to slow the race when the people building the systems say the risks are rising?
Watch Desk analysis
What happened
The researchers’ video testimonials were collected by AI safety nonprofit Palisade Research for frominside.ai and shared exclusively with Reuters, according to Rappler’s report on 5 October. Neel Nanda, a research scientist at DeepMind, said he believed there was at least a 10% chance AI could cause human extinction. OpenAI alignment research engineer Juan Felipe Ceron Uribe described frontier labs as racing “kind of blindfolded”.
The project’s contributors also argue that companies reward employees for building new models more than for urging caution. These are researchers’ warnings and judgements, not evidence that self-improving systems are already here or that catastrophe is inevitable.
Why it matters
The debate is not only about what future systems might do. It is also about whether people inside AI labs can influence decisions, and whether the public hears their concerns beyond the usual conference panels and social feeds. The testimonials offer unusually direct accounts from people who have worked on the technology, while leaving the scale of the risks contested.
Our read
A stated probability of human extinction deserves scrutiny, not a reflexive nod or an eye-roll. The useful contribution here is the researchers’ account of both technical fears and workplace incentives: safety depends partly on whether caution has any standing when a lab is racing competitors. A video project can make that argument visible; it cannot settle the probability question on its own.
What to watch
- Whether the project publishes further accounts and identifies contributors’ roles and evidence.
- How the named AI companies respond to claims about safety priorities and internal incentives.
- Whether policymakers or labs propose concrete steps to assess risks from self-improving systems.
Discussion spark: What evidence should persuade AI labs to slow development: a demonstrated technical risk, or credible warnings from researchers working inside the field?
Sources and evidence
- AI researchers warn companies rushing self-improving systems despite safety risks – Rappler (5 October 2026, 10:55 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.