Claude Mythos Preview Zero-Days · Meta Muse Spark · US-China AI Confrontation

12-hour AI industry digest

Top Stories

1. Anthropic Releases Claude Mythos Preview: Cybersecurity AI Discovers Thousands of Zero-Day Vulnerabilities

Date: April 8
Source: The Hacker News, TechCrunch

Anthropic unveiled Claude Mythos Preview, an advanced AI model designed specifically for cybersecurity that has discovered thousands of previously unknown zero-day vulnerabilities in major systems. The model is deployed through “Project Glasswing” with limited partnerships established with over 40 companies including Microsoft, Amazon, Apple, Google, NVIDIA, CrowdStrike, and Palo Alto Networks, for defensive security purposes only.

Due to the model’s unprecedented ability to identify software vulnerabilities, Anthropic is restricting access to prevent malicious use by cybercriminals. The company has been in ongoing discussions with U.S. government agencies including CISA regarding the model’s cybersecurity capabilities.

Impact: This marks a significant turning point in AI safety—the first AI model restricted from public release for being “too powerful,” sparking global discussion about balancing AI capabilities with security.

2. Meta Releases Muse Spark: Closed-Source Model Ranks Fourth in AI Benchmarks

Date: April 8
Source: The Next Web, CNBC, Bloomberg

Meta released Muse Spark, the first major large language model developed by Meta Superintelligence Labs led by Chief AI Officer Alexandr Wang. The closed-source model ranks fourth on the Artificial Analysis Intelligence Index v4.0 with a score of 52, behind Gemini 3.1 Pro Preview and GPT-5.4 (both 57) and Claude Opus 4.6 (53).

Muse Spark excels in chart understanding (CharXiv Reasoning 86.4%) and medical reasoning (HealthBench Hard 42.8%), but performs weaker on abstract reasoning tasks (ARC AGI 2 only 42.5). The model uses parallel sub-agent architecture, supports “thinking mode” for complex tasks, and achieves 77.4% on software engineering benchmarks (SWE-bench Verified).

Impact: Meta continues catching up in the AI race, but the model still lags behind leaders in general reasoning capabilities.

3. US-China AI Confrontation Escalates: US Companies Accuse China of “Distillation” Technology Theft

Date: April 7-8
Source: LA Times, RoboRhythms

Major US AI companies including OpenAI, Google, and Anthropic are sharing intelligence, accusing Chinese companies of using “distillation” technology to extract capabilities from US AI models. Anthropic has explicitly blocked China-controlled companies from using Claude, pointing to three Chinese AI labs—DeepSeek, Moonshot, and MiniMax—as suspected of illegally extracting model capabilities.

This practice involves large-scale data requests to extract and reverse-engineer AI model capabilities, threatening to “surpass any single company or region,” constituting a national security risk. Distilled models typically lack safety guardrails against malicious use.

Impact: US-China AI technology competition enters new phase, potentially affecting global AI collaboration and open-source ecosystem.

Model & Technical Progress

4. Anthropic Revenue Surges to $30B Annualized Rate, Signs Massive Compute Deals

Date: April 7
Source: TechCrunch

Anthropic significantly expanded compute agreements with Google and Broadcom to address explosive Claude model demand growth. Company revenue annualized rate surged from $9B at end of 2025 to $30B.

The company now serves over 1,000 enterprise customers spending over $1M annually, recently completing $30B Series G funding round at $380B valuation. Despite being listed as a supply chain risk by the US Department of Defense, this massive compute agreement represents Anthropic’s largest infrastructure investment to date.

Highlight: Anthropic grew from startup to AI giant within 12 months, with $30B revenue annualized rate keeping OpenAI executives “up at night.”

5. Zhipu AI Open-Sources GLM-5.1 While Raising API Prices

Date: April 8
Source: Planet News

Chinese AI company Zhipu AI open-sourced its flagship model GLM-5.1 while raising API prices by 10%. This is the company’s second price increase this year, following a comprehensive reform in February that raised coding subscription plan rates by over 30%.

This strategic move marks a significant shift from open-source accessibility to commercial sustainability as Chinese AI companies intensify competition with US rivals. GLM-5.1’s open-source release coincides with China accelerating its comprehensive AI strategy through multiple coordinated initiatives.

Impact: Chinese AI enterprises are exploring balance between open-source and commercialization to maintain advantage in global competition.

Healthcare AI Breakthrough

6. University of Oxford Develops AI Tool: Predicts Heart Failure Risk 5 Years in Advance

Date: April 8
Source: Planet News

Oxford University scientists achieved a major medical breakthrough, developing an AI tool that can predict heart failure risk 5 years before symptoms appear. The technology demonstrated 86% accuracy in a comprehensive study of 72,000 patients in England.

Over 60 million people worldwide suffer from heart failure, a condition where the heart cannot pump blood effectively. Identifying cases before developing into full heart failure represents a significant advance in preventive medicine, enabling doctors to prepare better management strategies or even prevent the condition entirely.

Impact: AI applications in preventive medicine make significant progress, potentially transforming global cardiovascular disease management.

Trend Watch

Key Trends in Past 12 Hours

1. AI Safety & Responsibility: Claude Mythos release sparked global discussion about “whether overly powerful AI should be public,” marking AI industry entering new responsibility phase. Companies need to find balance between capability openness and safety protection.

2. US-China AI Competition Heats Up: From model distillation accusations to open-source strategy adjustments, US-China AI competition has expanded from technical level to national security and geopolitical level.

3. Healthcare AI Practicalization Accelerates: Oxford heart failure prediction tool shows AI moving from experimental to clinical utility in healthcare, with preventive medicine becoming new focus.

4. Enterprise AI Continues to Explode: Anthropic’s $30B revenue annualized rate proves strong enterprise AI demand, with AI shifting from experimental tool to core business infrastructure.

5. Open-Source & Commercialization Balance: Zhipu AI’s “open-source + price increase” strategy reflects new exploration by AI companies balancing open-source ecosystem with commercial sustainability.

Upcoming Watch

• OpenAI GPT-5.5 (codename Spud): Expected release in coming weeks, may bring new reasoning capability breakthroughs
• DeepSeek V4: In gray testing, new vision version and “expert mode”launching soon
• EU AI Act Full Implementation: Shifting from draft to execution phase, may affect global AI deployment strategies
• Global Semiconductor Shortage: Memory chip prices surged 6x, expected to continue until 2027 when new factories come online

Briefing generated: 2026-04-09 18:00 (Asia/Shanghai)
Coverage period: 2026-04-09 06:00 — 2026-04-09 18:00
Data sources: Public news reports, industry analysis, official announcements