US officials confirmed Alibaba's Qwen lab orchestrated the largest known AI distillation attack, logging 28.8 million fraudulent exchanges against Anthropic between April 22 and June 5.
Officials say Alibaba's Qwen lab ran the largest known AI distillation attack, logging 28.8 million fraudulent exchanges against Anthropic. The state-level accusation escalates an AI security cold war.
Twin quakes that struck Venezuela on June 25 have killed 589 people with an estimated 4,300 injured. The UN warns tolls could exceed 900 as hospitals near collapse.
Liquid AI released LFM 2.5 on June 26, a compact 230M-parameter model built on state-space and liquid neural networks. The open-source model reaches parity without a transformer backbone.
US officials confirmed that Alibaba's Qwen lab orchestrated what officials called the largest known AI distillation attack against Anthropic, logging 28.8 million fraudulent exchanges between April 22 and June 5, 2026. The campaign targeted Anthropic's Mythos model, focusing on software engineering and agentic...
Continue reading arrow_outwardUS officials have publicly blamed China's Qwen lab for the attack on Anthropic, turning a security incident into a geopolitical flashpoint.
US forces struck Iranian targets after Tehran violated the recent ceasefire, reigniting a conflict that markets hoped had cooled.
Sentiment sits in Extreme Fear as fresh Iran strikes and a 500-plus death toll in Venezuela rattle risk appetite.
Liquid AI's 230M-parameter LFM 2.5 abandons the transformer architecture entirely, betting that smaller and different beats bigger.
US officials naming the Qwen lab as the source of the Anthropic attack marks a shift from private breach disclosures to state-level accusation. Expect retaliatory framing from Beijing and pressure on US firms to harden model supply chains. The line between AI research competition and national security is now openly contested.
OpenAI is building an advertising business around ChatGPT while setting GPT-4.5's retirement for June 27. The dual move signals a company optimizing for revenue and consolidating its model lineup ahead of its public market ambitions. Ad-supported chat raises fresh questions about data use and answer neutrality.
The Supreme Court cleared the way to deport Temporary Protected Status holders, stripping protections and exposing affected residents to faster removal. The decision sides with the administration over immigrant advocates. It reshapes the legal footing for hundreds of thousands who had relied on TPS shields.
The Trust Tax
Read full analysis arrow_outwardUS officials confirmed that Alibaba's Qwen lab orchestrated what officials called the largest known AI distillation attack against Anthropic, logging 28.8 million fraudulent exchanges between April 22 and June 5, 2026. The campaign...
Liquid AI released LFM 2.5 on June 26, 2026, a compact 230M-parameter model built on a non-transformer architecture combining state-space and liquid neural networks. The open-source model reaches parity with transformer models 3 times...
At its first appearance at Cannes, OpenAI declared advertising a core part of its strategy, citing more than 900 million weekly active ChatGPT users with roughly one-fifth of queries expressing direct commercial intent. The company says...
OpenAI is retiring GPT-4.5 from ChatGPT this week, after issuing a 30-day notice, though the model remains available via API. The o3 model is scheduled to follow with retirement on August 26, 2026, continuing OpenAI's rapid model...
The US military launched new strikes inside Iran on June 26, targeting radar, missile, and drone storage sites after accusing Tehran of launching at least four drones in the Strait of Hormuz, one of which struck a Singapore-flagged...
Twin earthquakes that struck Venezuela on June 25 have reportedly killed 589 people, with an estimated 4,300 injured and hospitals near collapse; the UN warns tolls could exceed 900 with tens of thousands unaccounted for. International...
In a 6-3 ruling on June 26, the US Supreme Court's conservative majority handed the Trump administration a win, allowing it to end Temporary Protected Status for roughly 350,000 Haitian and 6,000 Syrian immigrants. The decision strips...
Israeli airstrikes killed two people and wounded a third in Markaba, southern Lebanon, on June 26, while Israeli soldiers bulldozed and burned homes in the town despite a ceasefire requiring Israel to end its occupation. Lebanese and...
A wildfire burning in Utah has been described by officials as the worst ever recorded in the state, prompting what authorities called a historic red flag warning across parts of the region. Severe weather and extreme fire risk are...
Vice President JD Vance praised former President Richard Nixon on June 26, characterizing the Watergate break-in and cover-up as a minor episode. The comments drew attention as an attempt to rehabilitate a presidency that ended in...
Audit your API exposure. Map every endpoint that could leak model behavior through distillation, then rate-limit and watermark outputs the way Anthropic now wishes it had before 28.8 million exchanges leaked.
Test a non-transformer model. Pull LFM 2.5 and benchmark its 230M-parameter state-space architecture against your edge workloads to see if you can cut inference cost without a transformer.
Migrate off GPT-4.5 now. With retirement set for June 27, repoint pipelines to a supported model today and log fallback behavior so production never depends on a deprecated endpoint.
Naming Qwen lab as the source of the Anthropic attack moves AI security from quiet disclosure to geopolitical conflict, and retaliation is now the base case. Meanwhile OpenAI chases ad revenue and retires GPT-4.5, while Liquid AI proves a 230M-parameter non-transformer can hold its own. The lesson is to assume your models will be probed, copied, and weaponized. Build with attribution, watermarking, and migration paths baked in from day one.
AI analysis, world news, markets, and tools. One briefing, delivered free.