Anthropic calls for global AI development freeze amid safety concerns
Consensus Summary
Anthropic, the world’s most valuable AI company, called for a global freeze on advanced AI development on June 5, 2026, warning that humanity risks losing control as AI approaches 'recursive self-improvement.' The company cited internal data showing over 80% of its codebase was written by its Claude AI, including fixes for 800 bugs in April 2026. Anthropic’s CEO, Dario Amodei, previously estimated a 25% chance of catastrophic outcomes. The Australian government, led by Assistant Minister Andrew Charlton, welcomed the call but has adopted a lighter-touch regulatory approach, abandoning mandatory guardrails for high-risk AI. Critics like PauseAI Australia argue governments—not companies—should dictate the terms of any freeze, while experts such as Steven Murdoch question whether Anthropic’s concerns represent a genuine shift or long-standing advocacy. The Guardian noted tensions between Anthropic’s safety rhetoric and its reported involvement in NSA offensive cyber operations, while SMH and The Age focused on the company’s commercial incentives and the challenges of coordinating a global slowdown.
✓ Verified by 2+ sources
Key details reported by multiple sources:
- Anthropic, the company behind Claude chatbot, called for a global freeze on the most powerful AI systems development on June 5, 2026 (published June 6 AEST in SMH).
- Anthropic's Mythos AI system was withheld from public release due to fears of devastating cyberattacks, with a staggered rollout planned to fix security flaws.
- Anthropic's valuation is approximately $US965 billion ($1.3 trillion), surpassing OpenAI (ChatGPT maker).
- Anthropic's CEO Dario Amodei, who left OpenAI over safety concerns, stated there is a 25% chance 'things go really, really badly' with AI.
- Australia's Albanese government welcomed Anthropic's call for a more considered approach to frontier AI development, with Assistant Minister Andrew Charlton leading Labor’s AI strategy.
- Anthropic cited 'recursive self-improvement' as a critical threshold where AI systems could design and build their own successors with minimal human input.
- Anthropic reported that over 80% of new code merged into its codebase was authored by Claude AI as of May 2026.
- Australia's AI Safety Institute is expected to open within months, with a memorandum of understanding signed with Britain’s AI Security Institute.
- A global EY survey found 68% of Australians are worried about losing control over decisions made by AI.
- PauseAI Australia, part of an international movement, welcomed Anthropic’s call for a pause but argued governments—not companies—should set the terms.
Points of Difference
Details reported by only one source:
- The Albanese government shelved plans for mandatory guardrails on high-risk AI in December 2025, opting for a lighter-touch model instead of an EU-style AI Act.
- Former industry minister Ed Husic, who was dropped from the ministry last year, had argued a piecemeal approach would leave dangerous gaps.
- Anthropic’s internal data showed Claude resolved over 800 software bugs in April 2026 that would have taken a human engineer four years to fix.
- The Albanese government shelved plans for mandatory guardrails on high-risk AI in December 2025, opting for a lighter-touch model instead of an EU-style AI Act.
- Anthropic is embedding engineers inside the National Security Agency (NSA) to help use its Mythos model for offensive cybersecurity operations, despite a legal battle with the Pentagon.
- Steven Murdoch, a professor at University College London, called Anthropic’s call for a pause 'not surprising' and noted the company’s past actions, including supporting US offensive cyber capabilities.
- Heidy Khlaaf, chief AI scientist at the AI Now Institute, described the Mythos announcement as 'a marketing post' due to its vagueness.
- Anthropic filed for an IPO that could value the company at $1 trillion.
Contradictions
Conflicting information between sources:
- The Guardian states Anthropic’s call for a pause is not new, while SMH and The Age frame it as a significant intervention.
- The Guardian suggests Anthropic’s Mythos announcement may be more hype than substance, while SMH and The Age treat it as a credible security concern.
- The Guardian implies Anthropic’s support for NSA offensive cyber operations contradicts its safety narrative, while SMH and The Age do not mention this contradiction.
Source Articles
Anthropic urges global freeze on AI as it warns of losing control
The tech giant has offered to halt work on more powerful systems on one condition: that its rivals in the US and China agree to stop at the same time.
Anthropic urges global freeze on AI as it warns of losing control
The tech giant has offered to halt work on more powerful systems on one condition: that its rivals in the US and China agree to stop at the same time.
Anthropic urges AI development ‘pause’ and conversation about risks
US firm says it will convene policymakers, in post detailing progress of its Claude model towards ‘recursive self-improvement’ Anthropic has floated the idea of a worldwide “temporary pause” on AI development – and said it was going to convene “policymakers” to discuss the dangers of advanced AI – in its latest release touting the capabilities of its products. In a long post on Thursday, Anthropic detailed the progress of its AI model, Claude, towards “recursive self-improvement” – that is, bein