AI safety debate: why tech giants’ warnings are dividing regulators

The AI safety paradox: why tech’s warnings may backfire

The past week has seen a surge in high-profile warnings about artificial intelligence from the very companies driving its development. On Saturday, Anthropic CEO Dario Amodei published an essay calling for a "slowdown" in advanced AI model deployment, arguing that current safety measures are insufficient. His intervention follows similar concerns from other industry leaders, including OpenAI executives, who have repeatedly flagged existential risks posed by unchecked AI progress.

Yet these warnings arrive at a curious moment. The UK government, like others in Europe and North America, is already grappling with how to regulate a technology that evolves faster than policy can adapt. Amodei’s call for caution comes just as his company announces a $31bn datacentre project in western Queensland—a move that will significantly expand Anthropic’s computational capacity and, by extension, its ability to train more powerful AI systems. The contrast between public concern and private expansion has not gone unnoticed by regulators.

"When a company warns about the dangers of its own product while simultaneously scaling up production, it creates a credibility problem," said a senior official at the UK’s Competition and Markets Authority, speaking on condition of anonymity. The official pointed to a pattern: tech giants issue dire predictions about AI risks, then lobby against regulations that would impose meaningful constraints on their operations. "It’s not that the risks aren’t real—it’s that the solutions being proposed often align suspiciously well with industry interests."

Australia’s datacentre gamble: jobs vs. energy demands

Anthropic’s Queensland project, announced by Premier David Crisafulli on Wednesday, promises to deliver "thousands of jobs" and inject renewable energy into the state’s grid. The company has leased a site in western Queensland, a region that has struggled with economic diversification since the decline of traditional industries. For local officials, the deal represents a rare opportunity to position Australia as a player in the global AI race.

But the environmental and economic trade-offs are already under scrutiny. Datacentres are notorious for their energy consumption, and while Anthropic has pledged to use renewable sources, critics question whether the grid can handle the additional load without relying on fossil fuel backups. A 2025 report from the Australian Energy Market Operator warned that Queensland’s renewable energy infrastructure remains "fragile," with frequent supply shortfalls during peak demand periods.

The project also raises questions about the long-term value of AI-driven job creation. While Anthropic has promised thousands of roles, the bulk of these are likely to be in construction and maintenance—positions that may not offer the same stability as the high-tech jobs initially advertised. "We’ve seen this before with mining and manufacturing booms," said Professor Sarah Bell of the University of Queensland’s School of Engineering. "The jobs arrive, but so do the energy costs and the environmental trade-offs. And when the boom ends, the community is left with infrastructure it can’t maintain."

For now, the Queensland government appears willing to take the risk. Crisafulli described the deal as a "major win" for the state, one that would "put Queensland on the map as a leader in the next industrial revolution." But as AI datacentres proliferate globally, the question remains: who truly benefits from these projects—the communities hosting them, or the companies building them?

Copyright in the age of AI: Australia’s opt-out dilemma

While the UK and EU have taken steps to clarify copyright rules around AI training data, Australia is considering a more radical approach. Leaked documents from the Albanese government reveal plans for an "opt-out model" that would allow AI companies to scrape publicly available content—including personal photos, small business websites, and creative works—unless individuals or organisations explicitly prohibit it.

The proposal has sparked outrage among artists, writers, and digital rights groups, who argue that it effectively reverses the burden of protection. "Under current copyright law, creators have the right to control how their work is used," said musician and activist Holly Rankin. "This proposal flips that on its head. Suddenly, every family photo you’ve ever posted online is fair game unless you take the time to opt out."

The government’s rationale is twofold: first, that an opt-out system would reduce legal uncertainty for AI developers, and second, that it would position Australia as a more attractive destination for tech investment. But critics counter that the move would disproportionately harm independent creators, who lack the resources to monitor and enforce opt-out requests. "This isn’t about innovation—it’s about giving tech giants a free pass to exploit the work of others," Rankin said.

The debate mirrors similar tensions in the UK, where the Intellectual Property Office has faced criticism for its handling of AI and copyright. Last year, the UK government proposed allowing AI companies to use copyrighted material for training without permission, only to backtrack after a backlash from the creative industries. Australia’s opt-out model, if implemented, could face similar resistance—and legal challenges.

What’s next: the regulatory tightrope

As governments scramble to respond to AI’s rapid evolution, the industry’s mixed signals are complicating the policy landscape. On one hand, tech leaders warn of existential risks; on the other, they continue to expand their operations at breakneck speed. For regulators, the challenge is distinguishing genuine concern from strategic posturing.

The UK’s approach has been cautious but proactive. Earlier this year, the government introduced the AI Safety Institute, tasked with evaluating risks posed by advanced models. Yet the institute’s mandate is limited to research and recommendations—it lacks the authority to enforce binding rules. Meanwhile, the EU’s AI Act, which comes into full effect next year, imposes stricter requirements on high-risk applications, but its impact on foundational models like those developed by Anthropic and OpenAI remains unclear.

For now, the most immediate test may come from Australia. If the opt-out model moves forward, it could set a precedent for other countries grappling with the same questions. But with public trust in tech companies at a low ebb, any policy perceived as favouring industry over creators—or communities over corporations—risks sparking a backlash that could derail the entire regulatory effort.

One thing is certain: the AI safety debate is far from over. And as the industry’s warnings grow louder, so too do the questions about who stands to gain.