Two of the most powerful forces shaping artificial intelligence — Anthropic and OpenAI — have stepped forward to warn the world about the very dangers their own creations may pose, while simultaneously offering themselves as the guides who should lead humanity through those dangers. It is an ancient tension: those who build the labyrinth proposing to draw the map. As governments around the world begin the slow work of writing rules for a technology they do not yet fully understand, the question of who holds the pen — and whose interests that pen serves — may prove as consequential as the rules
AI Giants Push for Safety Standards While Shaping Regulatory Future
Companies proposing the rules that govern their own industry
Mark
So these companies are saying AI is risky, but they're also the ones writing the safety rules. That seems backwards.
Mimi
It's more complicated than that. They do have real expertise about how their systems work and what could go wrong. But yes, there's an obvious conflict when the people building the technology also get to propose how it's regulated.
Luke
The article doesn't actually tell us what specific proposals they've made or how they differ from what independent researchers or other companies are suggesting. We know they're advocating for frameworks, but we don't know what those frameworks actually contain.
Mark
Fair point. So we can't tell if their proposals are self-serving or genuinely good for safety?
Mimi
Not from this reporting, no. We'd need to see the actual proposals side by side with what other voices are recommending. That's the real test.
Luke
And we don't know how much influence they're actually having. Are policymakers listening to them? Are they pushing back? Are other companies proposing different approaches?
Mark
So the story is really about a potential problem, not a confirmed one.
Mimi
Right. The problem is structural—companies with financial interests are shaping rules that affect their own interests. Whether that actually produces bad outcomes depends on whether regulators have the independence and expertise to push back.
Luke
Which we also don't know yet. This is early. The frameworks are still being written.
Mark
Then what should someone watching this actually pay attention to?
Mimi
Watch whether independent oversight mechanisms get built. Watch whether multiple companies get a voice, not just the biggest ones. Watch whether regulators hire their own technical experts or stay dependent on what industry tells them.
Luke
And watch what the actual rules say when they're finally written. That's where you'll see whether this positioning mattered.
O Pulso
- Anthropic and OpenAI are publicly sounding alarms about AI risks while quietly drafting the regulatory blueprints they want governments to adopt — a dual move that is as strategically shrewd as it is openly visible.
- The tension lies in incentive: companies proposing their own oversight frameworks have every reason to design rules rigorous enough to reassure the public but lenient enough to preserve their own competitive edge.
- Smaller AI competitors and independent researchers risk being sidelined if compliance costs and governance structures are shaped primarily by the two firms already dominant in the field.
- Policymakers are caught in a genuine bind — they need technical expertise to regulate a technology they don't fully grasp, and the companies building it are the most available source of that knowledge.
- The critical variable now is whether regulators treat industry proposals as a starting point or a conclusion — and whether independent oversight bodies emerge with the capacity to push back.
Two of the most powerful forces shaping artificial intelligence — Anthropic and OpenAI — have stepped forward to warn the world about the very dangers their own creations may pose, while simultaneously offering themselves as the guides who should lead humanity through those dangers. It is an ancient tension: those who build the labyrinth proposing to draw the map. As governments around the world begin the slow work of writing rules for a technology they do not yet fully understand, the question of who holds the pen — and whose interests that pen serves — may prove as consequential as the rules themselves.
Two of the world's most influential AI companies are doing something striking in plain sight: warning about the dangers of the technology they are building, while positioning themselves as the natural architects of the safeguards meant to contain it. Anthropic and OpenAI have become prominent voices for AI safety and regulation — but their advocacy extends beyond raising alarms. They are actively proposing which agencies should hold authority, what oversight mechanisms should exist, and how compliance burdens should be distributed across the industry.
This is not unprecedented. Companies routinely participate in regulatory design, and their technical knowledge can be genuinely useful. Both firms employ leading AI safety researchers, and their understanding of what is technically feasible carries real weight. But participation and authorship are different things, and the line between them matters enormously here. A company that drafts regulatory frameworks has an obvious interest in frameworks that satisfy public concern without undermining its own competitive position — or that raise barriers high enough to disadvantage smaller rivals.
The stakes are considerable. AI governance is still being written in most of the world, and the frameworks taking shape now will likely define the industry for years. If Anthropic and OpenAI successfully shape those frameworks, they may entrench advantages, set compliance costs that only large incumbents can absorb, or secure favorable interpretations of ambiguous rules. If their proposals are genuinely well-designed, they could instead establish safety standards that meaningfully reduce harm.
Policymakers face a real puzzle with no clean solution. Excluding industry voices would be both impractical and counterproductive. But accepting their proposals without independent scrutiny would be equally unwise. The path forward likely requires regulators to build their own technical capacity, invite a broader range of industry voices into the conversation, and ensure that oversight mechanisms include actors with no financial stake in the outcome. Whether that kind of governance actually emerges — or whether the industry ends up largely writing its own rules while welcoming the appearance of oversight — remains the open question.
Two of the world's most influential artificial intelligence companies are simultaneously sounding alarms about the risks their own technology poses and positioning themselves as the architects of the safeguards meant to contain those risks. Anthropic and OpenAI have become vocal advocates for AI safety standards and regulatory frameworks, framing themselves as responsible stewards willing to accept oversight—while their public campaigns for specific governance structures could substantially shape which rules actually get written and who gets to enforce them.
The dynamic is striking in its transparency and its tension. Both companies have published detailed safety concerns about advanced AI systems, warning of potential harms that range from economic disruption to more severe risks. These warnings are not hypothetical; they describe challenges that emerge directly from the systems these companies have built and continue to develop. By raising these alarms publicly, Anthropic and OpenAI are signaling to policymakers, investors, and the public that they take the stakes seriously. They are also, implicitly, establishing themselves as credible voices in conversations about what should be done.
But the second part of their strategy complicates the first. As these companies advocate for safety measures, they are simultaneously proposing the specific regulatory architectures that would govern their industry. They are suggesting which agencies should have authority, what kinds of oversight mechanisms should exist, and what compliance burdens should fall on companies of different sizes. In other words, they are not simply warning about dangers; they are drafting the blueprints for how those dangers should be managed—and by extension, who should manage them.
This is not necessarily sinister. Companies often participate in regulatory design, and their technical expertise can be genuinely valuable. Anthropic and OpenAI do employ some of the world's leading researchers in AI safety. Their input on what is technically feasible or what oversight mechanisms might actually work is worth hearing. The problem, from a governance perspective, is one of incentive alignment. A company that proposes regulatory frameworks has an obvious interest in proposing frameworks that are stringent enough to satisfy public concern but not so stringent that they become uncompetitive or that they disadvantage the company relative to potential rivals.
The stakes of this positioning are substantial. AI regulation is still in its infancy in most jurisdictions. The frameworks being discussed now—in the United States, the European Union, and elsewhere—will likely shape the industry for years. If Anthropic and OpenAI successfully influence those frameworks, they could entrench their own competitive advantages, establish compliance costs that smaller competitors cannot afford, or secure favorable interpretations of rules that remain ambiguous. Alternatively, if their proposals are genuinely well-designed and adopted, they could establish safety standards that actually reduce risks.
Policymakers face a genuine puzzle. They need technical expertise to write sensible rules about technology they do not fully understand. The companies building that technology are the most obvious source of that expertise. Yet those same companies have financial incentives that may not align with the public interest. The solution is not to exclude industry voices from regulatory discussions—that would be both impractical and unwise. The solution is to ensure that industry proposals are tested against independent analysis, that multiple companies have a voice in shaping rules rather than one or two, and that oversight mechanisms include actors outside the industry itself.
What happens next will depend partly on how regulators respond to the safety initiatives Anthropic and OpenAI are proposing. Will they adopt the companies' frameworks wholesale, or will they treat them as one input among many? Will they establish independent technical capacity to evaluate industry claims, or will they rely primarily on what companies tell them? The answers to these questions will determine whether AI governance becomes a genuine partnership between industry and public interest, or whether it becomes a process in which the industry largely writes its own rules while claiming to welcome oversight.
Quer a matéria completa?
Leia o original em Associated Press ↗