The builders are drawing their own rules
They are building the future. They are also building its cage. The world’s leading artificial intelligence companies, normally locked in a furious race for supremacy, are talking to each other. They are discussing a truce. OpenAI, Google and Anthropic, three American firms that dominate the field, confirmed on Tuesday that they are planning a new body to set safety standards for their own technology. The stated goal is to manage the risks of the powerful systems they are creating before those systems can cause widespread harm, a remarkable act of self policing from an industry famous for its ‘move fast and break things’ ethos. They are rivals. Now they are collaborators.
Confirmation came on 15 September. It was not sudden. Reports confirm these talks have been taking place quietly for weeks, far from public view. The participants represent the industry’s vanguard. OpenAI, maker of the viral chatbot ChatGPT, has become a household name and a key partner of Microsoft. Beside it sits Google DeepMind, the formidable research division of Alphabet, a corporation whose technology is embedded in the lives of billions. The third party is Anthropic, a smaller firm founded by former senior members of OpenAI specifically to pursue safer development. These are the architects.
The companies are responding to rising pressure. They say so themselves. A statement from OpenAI cited growing concerns over the risks of the technology as a reason for greater oversight, an unusual admission of anxiety from the very people building the tools in question. The move can be read as an attempt to get ahead of regulation, proposing their own set of rules before politicians in Westminster or Washington impose stricter, and perhaps less favourable, ones. By creating their own standards body, the incumbent giants could define what ‘safe’ AI looks like, framing the debate for years to come and potentially creating new barriers for any future competitors who lack their vast resources. The rules are unwritten. They want to hold the pen.
What are they so afraid of?
The creators are afraid. They are haunted. The anxiety from within the industry is not about business rivals or flagging share prices, but a much deeper, more elemental fear that the tools they are building could escape their control and inflict irreversible damage on a global scale. This is not fiction. The concerns are concrete. Leaders at these firms have spoken publicly about the potential for their technology to enable mass disinformation that could shatter democratic processes, or to help rogue actors design novel pathogens. They worry about AI powered cyberattacks overwhelming critical infrastructure. They fear their work.
Beyond these immediate threats lies a more profound problem. The problem is control. The central anxiety of the AI safety field is that researchers are racing to build systems that will eventually become far more intelligent than their human creators, without any reliable method for ensuring those systems remain benevolent. That is the gamble. Anthropic was born from this. It was founded by people who left OpenAI because they believed the race for capability was dangerously outpacing the science of safety. Some developers have compared their work to the Manhattan Project. They are building a world changing weapon of immense power, but this time nobody is quite sure how the atomic chain reaction actually works.
The technology's behaviour is unpredictable. Even its creators do not fully understand the inner workings of their most advanced models, which learn and evolve in ways that defy simple explanation. They are black boxes, so they propose rules. The companies portray this collaboration as an act of sober responsibility, a necessary attempt to install guardrails on a road that is getting faster and steeper. Progress has been explosive. The capabilities of these systems have leaped forward in just the last few years, a rate of change that has surprised even the experts building them. This rapid scaling is the engine of their commercial success. It is also the source of their dread. They are the high priests of a new technology, warning that the god they are summoning might be a demon, and are now frantically trying to write the scripture that could bind it.
Washington has other ideas
But these fears do not resonate in Washington. They have a different priority. The American government is not primarily concerned with the philosophical dangers of superintelligence. It is concerned with China. For policy makers, the development of artificial intelligence is not a collaborative scientific project to be managed for the good of humanity. It is a zero sum race for global dominance. This is the central conflict. The companies talk of brakes. The state demands an accelerator.
A report from TechCrunch on 15 September confirmed this schism. It noted that talks between the tech giants have been happening for weeks. It also noted the view of the Trump team. They dismiss the safety worries. Their focus is simple. Keep pace with China. This perspective reframes the entire debate, shifting it from a careful discussion about managing risk into a blunt geopolitical imperative where the greatest danger is not an uncontrollable superintelligence but falling behind a strategic adversary. In this new cold war, slowing down for safety looks like unilateral disarmament. It is seen as a strategic folly.
The fear in the West Wing and the Pentagon is not that a powerful AI might one day escape human control, but that a powerful AI controlled by the Chinese Communist Party might arrive first, giving Beijing an insurmountable economic and military advantage. This is not a theoretical exercise. It is a matter of national security. The creators of this technology might be consumed by existential dread and the complex ethics of their work, but American political leaders see a far simpler calculation driven by national interest and a stark rivalry with Beijing. That calculation prizes speed above all else.
This leaves the new safety consortium in an impossible position. OpenAI, Google and Anthropic are all American companies. They operate within a system that now views their work through a geopolitical lens. Any self imposed rules or ethical guardrails they might agree upon will be seen as a potential handicap in the race against a competitor that is not bound by the same constraints. The government wants supremacy. The builders want survival. Washington’s pressure to innovate faster, to deploy quicker and to worry later could render any industry led safety initiative irrelevant before it even begins.
Beijing is not in the room
The consortium has a fundamental, perhaps fatal, weakness. Its architects are all American. OpenAI, Google, Anthropic. They are companies born of Silicon Valley, operating under United States jurisdiction and embedded deeply within its national innovation system. Beijing is not in the room. This is not an oversight. It is a reflection of the geopolitical reality that shapes the entire field of artificial intelligence, a reality defined by intense competition between Washington and Beijing. Any body formed without China’s participation is, by definition, not a global standards body but a Western one, an alliance of pioneers whose influence ends precisely where their greatest competitor’s begins. The initiative is crippled from the start.
A safety framework for artificial intelligence only works if it is universal. The potential risks do not respect national borders. They are inherently global. But the proposed rules are not. This geographical and political limitation ensures that any standards emerging from the OpenAI and Google talks will be interpreted in Beijing not as a sincere effort to manage existential risk, but as a strategic manoeuvre. It will look like an attempt by incumbents to set the rules of the game. It will be seen as an effort to codify a Western technological advantage, using the language of safety to create a regulatory moat that protects American firms from international competition. Nothing more.
This structural absence fuels the very anxieties the body seeks to calm. It validates the central argument of American hawks. Their position is simple. Why should the United States bind its most innovative companies with safety rules when its primary adversary will do no such thing? The sight of a Western only safety club reinforces the perception of a dangerous two track world, one where American labs proceed with caution while Chinese labs race ahead without similar constraints. This is the core dilemma. The creation of a limited safety pact might inadvertently provide the most compelling political argument for abandoning the pursuit of safety altogether in favour of a desperate, unregulated sprint for primacy.
Ultimately, the group’s composition defines its ceiling of achievement. It is a club for participants who already share a broad, if contested, consensus on the nature of the problem and the importance of democratic oversight. It is a Western solution. For a global problem. Its pronouncements, however thoughtful, will echo only within its own chamber, unable to reach, let alone persuade, the engineers and state planners in Beijing who are its intended but absent audience. The body can set standards for itself. It cannot set them for the world. That is its weakness.
Who really benefits from these rules?
The companies say this is about safety. An altruistic mission. This is a convenient argument for a group of corporations who benefit enormously from being seen as responsible actors in a field that many people find frightening. The rules they propose, born from these private talks, could easily become a moat. A very deep one. The kind of safety and testing protocols a body led by OpenAI, Google and Anthropic might demand would almost certainly involve immense computational power, creating an economic barrier that only they themselves could clear. Frontier models already cost hundreds of millions of pounds to train. The addition of complex, mandatory safety evaluations could push that price towards a billion pounds for a single model, a sum far beyond the reach of any university lab or fledgling start up. This is not philanthropy. It is strategy.
The pattern is familiar. A known playbook. Technology incumbents have long sought to write the rules that govern their own markets, framing commercial self interest as a matter of public good. Microsoft faced exactly these accusations over its practices in the 1990s. The playbook involves using a dominant market position to shape regulation, locking out future competition under the guise of ensuring stability or protecting consumers from harm. The founders of these AI labs are creating a rulebook for a game they are already winning, a move that could cement their market position for the next decade. Sam Altman, the chief executive of OpenAI, has spent more than a year touring world capitals to advocate for exactly this kind of regulatory framework. He is lobbying. This is a blueprint for a cartel.
A new standards body offers these giants legitimacy. It buys them public trust. It provides them with a social licence to operate from governments that are increasingly nervous about the technology’s trajectory. By positioning themselves as the sole arbiters of what is safe, they can continue their own research with reduced political interference while smaller, more agile competitors become entangled in costly compliance regimes. It is a clever defence. It placates the public. It strangles new rivals. The discussions happening in the offices of Palo Alto and Mountain View are not just about preventing hypothetical AI disasters. They are also about preserving market structure. They are about who gets to build the future, and on whose terms. The new body could become the ultimate gatekeeper, deciding which models are safe enough for public use, a judgment made by the very companies that have the most to lose from disruptive innovation. This is power. It is not safety. It is control.
The limits of self-regulation
The new body will likely form. It is the logical next step. Its creation will be announced with solemn statements about responsibility and the careful stewardship of powerful technology. But its founding charter will immediately collide with the immovable objects of national interest and international rivalry. The body's power will be constrained from its first day. Its writ will not run far. The real world has other plans. The true decisions that will shape the deployment of artificial intelligence will not be made by this committee, but by governments in Washington and Beijing who see the technology not as a shared global resource but as a critical asset in a new great power competition.
Washington has other priorities. The source of the most significant headwind will be the very government in whose jurisdiction these companies operate. An American administration, particularly one shaped by the thinking of Donald Trump’s team, views the AI race through a simple, brutal lens. It is a race against China. The overriding imperative is to win it. Any industry led initiative that threatens to slow down American progress, even in the name of safety, will be seen as a strategic own goal. A voluntary pause will be viewed as a unilateral disarmament. The pressure from the government will not be to be safe, but to be first. The standards body may publish its recommendations, but it cannot compel the Pentagon or the White House to listen if they believe national security requires them to move faster.
This consortium has a fatal flaw. Its membership is exclusively Western. China is not in the room. The standards agreed in Palo Alto will hold little sway over the research priorities being set in the laboratories of Shanghai or the policy directives issued from the highest levels of the Chinese Communist Party. China's AI development proceeds along a parallel track, governed by its own internal logic and its own national goals. An industry body that excludes one of the two main players in the field cannot claim to be setting a global standard for safety. It is setting a regional one. The result will be a fractured world, with two distinct AI ecosystems developing according to different rules, a situation that makes systemic risk far harder to manage.
So what should we expect? The new body will exist. It will produce reports, host conferences and offer technical benchmarks that could become de facto standards for Western firms. This will not be meaningless. It will provide a forum for collaboration and a vocabulary for discussing risk. It will be a start. But its real world impact on the global trajectory of AI development will be marginal. It cannot stop the geopolitical race. It cannot bind the actions of a determined state competitor. The safety protocols it devises will be perpetually vulnerable to being overridden by the demands of national governments obsessed with maintaining a technological edge. The committee will have authority, but it will not have power. The fate of this technology is, for now, out of its hands.
Sources. France 24: OpenAI, Anthropic and Google in talks to create a standards body as AI fears rise. TechCrunch: OpenAI, Anthropic, Google have been in talks on AI safety for weeks.
Analysis. Drafted with AI assistance from the sources listed above and reviewed by an editor before publication. Jnews links to the organisations it writes about.
