Anthropic Suspends Two Models Amid Concerns Raised by Amazon CEO
On June 13, Anthropic suspended access to its new models, Claude Mythos 5 and Claude Fable 5. Both models are currently listed as "Monitoring" on the company's status page, and the move is described as a response to U.S. government security concerns.
What is drawing even more attention is the apparent source of those safety concerns. According to TechCrunch, Amazon CEO Andy Jassy may have relayed security concerns about Anthropic's models to government agencies. Amazon occupies a complex position as both a major investor in Anthropic and its cloud infrastructure provider, so the report's impact on the industry is not insignificant.
The company continues monitoring with the aim of restoring the models, though no timeline for their return has been given.
KPMG Retracts AI Adoption Survey Report Over Hallucinations
KPMG, one of the Big Four accounting firms, has withdrawn a survey report on AI adoption shortly after publishing it. The reason: the report allegedly contained AI-generated hallucinations — plausible-sounding falsehoods. It is an ironic turn of events, with a report about AI becoming a case study in AI's reliability problems.
The episode underscores, once again, how critical fact-checking is when companies publish AI-assisted content.
UK Police Officer Suspected of Fabricating Evidence with AI — Under Investigation Across Multiple Cases
A police officer with Derbyshire Constabulary in the UK is under investigation on suspicion of having "created evidence" in multiple cases using AI. Revealed by Sky News, the case has also drawn attention on Hacker News, picking up 19 points.
Because misjudgments in law enforcement AI use can directly infringe on individual rights, governance has become an urgent issue. The case stands as a symbolic example of the dangers of treating AI-generated content as evidence.
China's Indium Phosphide Export Restrictions Threaten AI Data Center Buildouts
According to Reuters, supply restrictions on indium phosphide, whose exports China controls, are beginning to affect AI data center rollout plans. Indium phosphide is essential as a material for optical communication components and is one of the foundational technologies underpinning faster, larger-scale AI infrastructure.
Export restrictions on scarce materials, following on from semiconductors, show that geopolitical risk is now extending to the physical supply chain of the AI race.
Jeff Bezos' AI Startup "Prometheus" Raises $12 Billion
Prometheus, the AI startup led by Amazon founder Jeff Bezos, has reportedly raised $12 billion (roughly ¥1.7 trillion). The company's valuation has reached $41 billion, and it aims to build a "general-purpose AI engineer" that operates in the physical world.
This massive investment in physical AI applications beyond the software domain signals a serious entry into fields such as robotics and manufacturing, where the benefits of AI have so far been limited.
Open-Source Model Watch: DeepSeek v4 Pro's "Cost-Performance" Problem and a New Model from Cohere
In the local LLM community, debate over the balance between model size and performance is heating up. DeepSeek v4 Pro is among the largest open models at 1.6 trillion parameters, yet observers point out that it often trails GLM 5.1 (750 billion parameters), Kimi K2.7, and MiniMax M3 (roughly 450 billion parameters) on many benchmarks. Others, meanwhile, emphasize the importance of its Huawei-based inference infrastructure.
Separately, Cohere has released North Mini Code, an efficient code-generation model with 30B-A3B (30 billion total parameters, 3 billion active), under the Apache 2.0 license, and architecture support has been added to llama.cpp. Featuring a 256K context length, the model is optimized for code generation and agentic software engineering.
---