According to Gate data, the US dollar index (DXY) continued to rise, breaking through 101 and reaching its highest level since May 2025.

In the past hour, total liquidations across the network reached $10.12 million, with $8.60 million from long positions, with $1.52 million from short positions. BTC liquidation $6.49 million, ETH liquidation $1.40 million.

According to Odaily Seer, the probability of Claude Fable 5 being restored for US customers by July 1st on Polymarket has risen to 73%, a 33% increase in 24 hours. If Anthropic reopens Claude Fable 5 (or Claude Mythos, or a version confirmed to be the same model) to the US public before the specified date, this event will be counted as "Yes"; otherwise, it will be counted as "No". Eligible restoration methods include public beta or a public waiting list; closed beta and private access are not counted. Other models such as Haiku, Sonnet, and Opus are not counted by default unless confirmed to be the same model as Claude Fable 5. The count is primarily based on Anthropic's official announcements, while also referencing mainstream media consensus reports. Due to export controls imposed by the US government on national security grounds, Anthropic urgently suspended global access to Claude Fable 5 a few days after its release on June 9th. While the restrictions primarily targeted foreign users, Anthropic ultimately closed access to all users, including those in the United States, due to the difficulty in verifying user identities in real time, and switched requests to weaker models such as Opus 4.8. Anthropic is currently in high-level communication with the White House to discuss issues such as model capabilities and potential jailbreak risks. The company has publicly stated that the ban may have stemmed from a misunderstanding and is working to restore access. Odaily Seer continues to monitor the prediction market, seeing changes before pricing.

Odaily reports that the Swiss Foreign Ministry has announced that the scheduled US-Iran talks, originally planned for Friday, will not proceed as planned. (Jinshi)

According to CoinWorld, Google DeepMind has released an AI control roadmap, aiming to design a system-level security framework for advanced AI agents deployed internally. This roadmap treats agents as potential "internal threats" and establishes a defense-in-depth system. The security system introduces trusted AI as a supervisory agent, monitoring the agent's inference chain, actions, and plans in real time, and directly intercepting any anomalies or malicious attempts detected. This defense mechanism draws on the MITRE ATT&CK adversarial framework, a standard in the cybersecurity industry, to establish a threat modeling system specifically for agents. The development team has analyzed 1 million task trajectories using a prototype system. Real-world data shows that the vast majority of anomalies that triggered warnings were not due to malicious intent, but rather stemmed from agents misunderstanding instructions or overloading their actions.