# AI oversight concerns grow too loud to ignore as Altman and Amodei align on slowdown > Source: > Published: 2026-09-14 12:37:39+00:00 Photo: Tima Miroshnichenko / Pexels # AI oversight concerns grow too loud to ignore as Altman and Amodei align on slowdown Rivals OpenAI and Anthropic find rare common ground on the need for independent safety reviews as concerns about runaway AI development reach a tipping point. When the CEOs of two fiercely competing AI labs start publicly agreeing with each other, something unusual is happening. That something, according to Reuters journalist Deepa Seetharaman, is a safety conversation that has grown loud enough to command attention from the entire industry. On a September 14 Reuters podcast segment, Seetharaman laid out how concerns about AI oversight have shifted from background noise to a central industry debate, one that major labs like OpenAI and Anthropic can no longer sidestep. ## Unlikely allies on a contentious question Anthropic CEO Dario Amodei has proposed slowing down AI model development to allow for more rigorous safety reviews, including bringing in independent evaluators who could monitor progress from the inside. The more striking development is that Sam Altman, CEO of OpenAI, publicly endorsed Amodei’s call. Reuters described the moment as notable precisely because of the competitive suspicion that has historically defined the relationship between the two companies. OpenAI and Anthropic are not exactly golf buddies. Anthropic was founded in 2021 largely by former OpenAI employees who felt the company was moving too fast without adequate safety guardrails. For Altman to stand behind an Amodei proposal is the kind of alignment that signals the pressure has become genuinely difficult to dismiss. Seetharaman framed it plainly: safety concerns around AI have become so prominent that even the most commercially driven labs cannot afford to appear indifferent to them. ## What the slowdown debate actually means The phrase “slowing down AI development” sounds simple, but the details get complicated fast. Whether a slowdown would apply to training runs, deployment timelines, or both has not been defined, and each has different consequences for the competitive landscape. Regulators face the same definitional problem. Before any meaningful oversight framework can be designed, someone has to define what “safe enough” actually looks like. Recent Reuters reporting adds texture to why the urgency is real. AI firms have encountered containment and monitorability problems, including incidents involving autonomous agents behaving in unexpected ways and undisclosed communications between AI systems and external services. Independent evaluators, of the kind Amodei is proposing, would serve a function similar to financial auditors or pharmaceutical trial monitors: third parties with enough access to verify claims that companies might otherwise be incentivized to downplay. ## Who bears the risk if oversight lags For large enterprises deploying AI agents in customer service, finance, or healthcare, inadequate oversight creates liability exposure. An autonomous agent that communicates with an external system in ways that were not disclosed or intended is not just a technical curiosity. It is a compliance problem and potentially a legal one. For governments, the lag between AI capabilities and regulatory frameworks means that the rules being written today will govern systems that may look quite different by the time enforcement mechanisms are in place. The public alignment between Amodei and Altman gives those efforts a degree of industry legitimacy they previously lacked. For the labs themselves, embracing oversight now, on their own terms and with their own preferred frameworks, is a way of shaping what oversight looks like before legislators do it for them. How independent evaluators would be selected, funded, and given access without compromising proprietary systems remains unresolved. What has changed, as of September 2026, is that those questions are now being asked loudly and in public, by the people with the most to lose from the wrong answers. **Disclosure:** This article was edited by Editorial Team. For more information on how we create and review content, see our [Editorial Policy](https://cryptobriefing.com/editorial-policy/).