AI Bosses Promise 'Moral' AI Safety, But Keep the Reins in Their Own Hands

Leaders of major AI companies have made a joint, non-binding pledge to put AI safety higher on the agenda. Notably, oversight of that pledge stays in-house, raising questions about its credibility.

AI safety is back at the top of the agenda now that a group of leading AI companies has made a joint, 'moral' commitment to ensure responsible development of artificial intelligence. The announcement, which surfaced in the tech press this week, confirms a pattern we've seen repeatedly in recent years: major players such as OpenAI, Google, Microsoft, Meta, and Anthropic solemnly promise to handle increasingly powerful AI models with care. What's new, however, is that the companies involved explicitly state they want to organize compliance oversight themselves, without an independent external watchdog.

What exactly does the pledge involve

The statement builds on earlier initiatives such as the voluntary safety commitments signed in the United States in 2023 and the agreements from the AI Safety Summits in Bletchley Park and Seoul. At its core are promises around testing frontier AI models for risks before release, sharing information about dangerous capabilities, and setting up internal 'red teams' that actively try to misuse models to expose weaknesses. There's also talk of transparency toward governments when a model turns out to have unexpectedly advanced abilities, for instance in the area of cyberattacks or biological risks. What's distinctive about this new round of commitments is the emphasis on self-regulation: the participating companies want to jointly determine what counts as 'safe enough,' rather than having that standard set by an independent body. Critics point out that this amounts to the butcher grading his own meat.

Which companies are leading the way

Although the exact makeup of the group varies by source, it's widely assumed that the biggest American AI developers are on board. Alongside OpenAI and Google DeepMind, Microsoft, Amazon, and Meta are also named as involved parties, while Anthropic has traditionally led the way in publicly backing safety frameworks. Chip giant NVIDIA is more often cited as an infrastructure partner than as a signatory to codes of conduct, and European players such as Mistral are also watching the discussion closely, although the emphasis of this particular agreement clearly lies with the American tech giants. This broad coalition speaks volumes: it shows that even fierce competitors recognize that public pressure and regulators can't be ignored. Anyone wanting to know more about how these companies relate to one another and how the sector has evolved can turn to the history of artificial intelligence, where the rise of these players is described in detail.

Criticism: who watches the watchmen

Not everyone is impressed. Researchers and advocacy groups point out that voluntary, unenforceable commitments have regularly been abandoned in the past as soon as commercial pressure increased. Without independent audits, fines, or legal obligations, a 'moral' pledge remains in practice primarily a PR tool, according to the criticism. Moreover, there's no clear mechanism for establishing when a company fails to live up to the agreements, let alone who can be held accountable for that. This discussion touches on a fundamental tension in the sector: on one hand, companies want to innovate and compete quickly; on the other, there's growing public demand for hard, legally anchored frameworks such as the European AI Act. While the EU is pushing for binding rules, American companies are, for now, opting for self-regulation — a difference in approach that may prove decisive in the coming years for what safe AI looks like worldwide.

What this means for users and businesses

For organizations that work with AI tools on a daily basis, this development mainly underscores one thing: trust in a vendor shouldn't be blind. It's wise to check for yourself what safety measures an AI provider actually applies, rather than relying on press releases alone. In AI applications we discuss how businesses can deploy AI responsibly, and our knowledge base offers practical tools for assessing risks yourself.

Conclusion

The joint pledge from the world's biggest AI players is a signal that the call for AI safety can no longer be ignored, but the choice of self-oversight simultaneously raises doubts about its effectiveness. Whether this promise holds up once commercial and geopolitical pressure increases remains to be seen in the coming months. Anyone wanting to follow developments closely can find more AI news on our site.

Vandaag en Morgen

---

Source: Vandaag en Morgen