Fable 5 launches publicly; AI now leads US job cuts
Anthropic's Fable 5 goes public with built-in guardrails, as AI-linked layoffs hit a 2026 record and Taiwan eyes chip export curbs.
By BINA Editorial
Today's dispatch covers a model release that splits AI access by safety tier, a chip-export policy shift in the Indo-Pacific, and a labour report with concrete numbers on AI's effect on employment — alongside the EU's revised compliance clock, a United Nations resource warning, and a legal contest over what the US government can demand of frontier AI companies.
Anthropic releases Claude Fable 5 alongside restricted Mythos 5
Anthropic made its most capable model publicly available on 9 June, releasing Claude Fable 5 — a version of its Mythos-class system with built-in safety guardrails — alongside the restricted Claude Mythos 5, available only to vetted partners, governments, and participants in Project Glasswing. Both models carry a one-million-token context window and a 128,000-token output limit, priced at $10 per million input tokens and $50 per million output tokens. The dual-track release separates general access from high-capability deployments: Fable 5 covers writing, coding, research, and image analysis; Mythos 5 unlocks cybersecurity and infrastructure-defence applications that require lifted restrictions. Amazon Web Services confirmed same-day availability on its Bedrock platform.
Taiwan moves to close the AI chip loophole with China
Taiwan's government is weighing export controls that would criminalise unauthorised AI chip sales to China for the first time, bringing its rules into closer alignment with US restrictions already in force. Under the proposed measures, restrictions would extend to all Chinese buyers — not only blacklisted entities such as Huawei — and would cover high-performance AI servers containing Nvidia chips. The move targets smuggling and third-party diversion routes that have enabled advanced hardware to reach China despite existing US bans. No legislative date has been set, but the proposal marks a significant shift in Taiwan's approach to semiconductor trade governance.
AI cited in nearly 40 percent of US layoffs so far in 2026
Artificial intelligence has overtaken all other recorded causes of job cuts in the United States during the first five months of 2026, according to a report by outplacement firm Challenger, Gray & Christmas. US employers announced more than 97,000 layoffs in May alone — the highest May figure since the early months of the COVID-19 pandemic — with AI cited as the primary reason for close to 40 percent of those cuts. The cumulative total of AI-linked layoffs in 2026 has reached 87,714, already surpassing the combined figures for both 2024 and 2025. Technology sector workers account for the largest share of affected roles.
EU extends AI Act deadlines and publishes watermarking rules
The European Commission has confirmed extended compliance deadlines for high-risk AI systems under the AI Act, while separately publishing draft guidelines on transparency and watermarking for generative AI. Standalone high-risk systems (Annex III) now have until 2 December 2027 to comply, and high-risk AI embedded in regulated products (Annex I) until 2 August 2028 — both pushed back from the original August 2026 target. Generative AI systems already on the market before August 2026 gain a four-month grace period for watermarking obligations, with a deadline of 2 December 2026. The guidelines specify that AI-generated content must carry machine-readable markings clearly identifiable as artificially produced; public consultation on the draft closed on 3 June.
UN flags AI's mounting resource footprint
A United Nations University study released 4 June projects that AI data centres could draw 945 terawatt-hours of electricity annually by 2030 — nearly three times the combined yearly consumption of Pakistan, Bangladesh, and Nigeria. Water use associated with AI infrastructure is on track to equal the basic domestic needs of 1.3 billion people by end of decade, while the physical land footprint of data centres could exceed 14,500 square kilometres. The study notes that day-to-day inference — not model training — accounts for 80 to 90 percent of total energy demand, placing the environmental cost squarely on widespread deployment rather than on model development alone.
Anthropic and the Pentagon face off in court over military AI contract
The Trump administration is defending against a lawsuit filed by Anthropic alleging unlawful retaliation after the Pentagon removed the company from classified military AI programmes. Anthropic contends the blacklisting followed its refusal to strip safety guardrails that prevent its models from supporting autonomous weapons or domestic surveillance systems. The Defence Department began formal testing of replacement models from OpenAI and Google in March 2026, with OpenAI subsequently securing a new contract. Court documents indicate the administration is contesting the retaliation claim; the case raises unresolved questions about what safety conditions the government may require — or forbid — in frontier AI procurement.