Researchers and former AI-lab employees call for a pause in the race toward self-improving systems
Current and former staff at OpenAI and Google DeepMind gave video testimonials to AI safety nonprofit Palisade Research, warning that their employers are doing too little to guard against the dangers of self-improving AI systems that could slip beyond human control. The project, called frominside.ai, was designed to let insiders speak directly to the public rather than through social media. Some participants said their concerns are genuine rather than a marketing posture.
Why it matters This is a notable break from the usual pattern of safety concerns being aired through op-eds or anonymous leaks: insiders at two of the most influential labs are going on record, via video, to warn about self-improving systems specifically. Coming amid intensifying competitive pressure between OpenAI, Google DeepMind and others to ship more capable agentic models, the testimonials add weight to calls for slower, more cautious development and could influence public and regulatory sentiment at a moment when trust in AI labs is already fragile.
What to watch Watch whether more employees go on record through frominside.ai, and whether OpenAI or DeepMind leadership issues any public response to the specific self-improvement concerns raised.