Rappler reports, citing Reuters, that current and former OpenAI and Google DeepMind researchers warn companies are doing too little to protect against the potentially disastrous fallout of self-improving AI systems that could outpace humans' ability to control them. In video testimonials collected by the AI safety nonprofit Palisade Research and shared exclusively with Reuters, the employees said their concerns about existential risk were sincere, not marketing, and that labs celebrate people building new models more than those urging caution. The project, called frominside.ai, is an attempt to share those concerns with the public beyond social media's echo chamber. "The risk is ramping up pretty fast," said Geoffrey Irving, co-founder and chief scientist at nonprofit Resolution, who has worked at both OpenAI and DeepMind.

Executives have tried to allay the concerns, but Rappler notes political pressure from President Donald Trump for US technology to keep an edge over China. Anthropic CEO Dario Amodei published an essay this month calling on the industry to slow down to "pace the frontier," and OpenAI CEO Sam Altman quickly concurred, yet both companies shipped new models this month competing for customers — though OpenAI said on Monday it had held back an even more powerful model. Irving pushed back on the framing: "If you're doing a very dangerous thing, you should just slow down. The AI companies are overplaying the extent to which this is a pure coordination problem." Rappler also reports Anthropic plans to caution IPO investors that advanced AI could pose "catastrophic or existential risks to humanity."

TechTarget's analysis argues that, for now, self-improving AI is less dramatic than it sounds — but it forces enterprises to get far more precise about what counts as a change. TechTarget cites Autoheal, whose Evaluator agent scores other agents' work and whose Healer proposes prompt, tool, skill or model changes, but only after testing against historical benchmarks, Git version control and engineer approval. Harness applies familiar delivery controls — canaries, risk scoring, rollback — to nondeterministic agents whose behavior can shift when a prompt or model setting changes. TechTarget's framing for CIOs: evidence should buy scope, not certainty. A prompt tweak improving an internal summary may justify broad use, while a small change touching refunds, ERP approvals or employment decisions may need to be broken into scoped, time-limited deployments with defined rollback paths.