US-Iran Direct Talks Begin in Islamabad as Hormuz Strait Traffic Remains at Bare Minimum
Source: https://www.163.com/dy/article/KQ7G9B8R05198NMR.html
US and Iranian delegations held their first direct negotiations on April 11 in Islamabad, Pakistan, led by US Vice President Vance. Trump said results would be clear within 24 hours, warning of intensified military action if talks fail. Iran has set two preconditions: a ceasefire in Lebanon and the unfreezing of Iranian assets.
The Strait of Hormuz continues to see traffic at less than 10% of pre-conflict levels, with only 4 vessels passing in the last 24 hours. Lebanon and Israel have agreed to discuss ceasefire arrangements for the first time at the US State Department on April 14.
My take: This is a high-stakes negotiation. Vance’s personal involvement signals strong US desire to exit the conflict, but Iran’s preconditions—involving multiple parties—make a comprehensive short-term deal unlikely. Oil prices plunged 14% this week on ceasefire optimism, but JPMorgan warns that if the Hormuz Strait doesn’t reopen until July, crude could rise another $15-20 per barrel.
Anthropic’s Mythos Model Sparks Wall Street Cybersecurity Panic; White House Urges Major Banks to Test
Source: https://www.cnbc.com/2026/04/10/coreweave-anthropic-claude-ai-deal.html
US Treasury Secretary Bessent and Federal Reserve Chair Powell this week convened an emergency meeting with CEOs of JPMorgan, Goldman Sachs, Citigroup, and Bank of America to discuss the cybersecurity threats posed by Anthropic’s latest model, Mythos. The model reportedly can identify and exploit vulnerabilities in mainstream operating systems and browsers, with regulators calling it the biggest new cyberattack risk facing the financial sector.
Mythos is currently available only to a select group of partners including Amazon, Apple, and JPMorgan. Canada’s central bank has followed suit, holding similar discussions with major financial institutions.
My take: This marks the first time AI capability has entered the highest levels of financial regulatory consciousness as a “security threat” rather than an “efficiency tool.” Anthropic’s decision to withhold its most powerful model on safety grounds has now forced regulators worldwide to take AI safety from academic debate to national security priority. The implications for the banking sector could be profound: institutions will need to fundamentally reassess whether their cybersecurity defenses can withstand AI-driven attacks.
Alibaba Confirms HappyHorse Video Generation Model, Tops Global Rankings Ahead of ByteDance’s Seedance 2.0
Alibaba’s ATH Business Group officially confirmed that the mysterious video generation model HappyHorse was developed in-house. The model anonymously topped the Artificial Analysis global text-to-video leaderboard with an Elo score of 1379, surpassing ByteDance’s Seedance 2.0, with its “image-to-video” subcategory setting a new all-time record on the leaderboard.
HappyHorse’s API will open on April 30. The news boosted Alibaba shares by over 4% intraday.
My take: Alibaba has finally revealed its hand in AI video generation—and it’s at the top of the global pack. This shifts the narrative that Alibaba was falling behind Baidu and ByteDance in the AI race. Video generation is one of the most promising AI application areas, and HappyHorse will now compete head-on with ByteDance, Runway, and Pika. Notably, this model comes from ATH, an internal innovation unit—suggesting Alibaba is building AI capabilities through internal incubation rather than acquisition.
DeepSeek V4 Rumored for Late April Release; Founder Confirms Internally
Source: https://www.stcn.com/article/detail/3740016.html
According to The Information and multiple Chinese media reports, DeepSeek’s next-generation flagship model V4 will be released in late April. Founder Liang Wenfeng confirmed this in internal communications. However, several AI entrepreneurs who have worked closely with DeepSeek caution against high expectations, suggesting V4 is unlikely to replicate the impact of V3.
DeepSeek’s web interface recently added “Fast Mode” and “Expert Mode” interaction options, which industry observers interpret as preparation for a more complete model lineup.
My take: DeepSeek V3’s launch in early 2025 was a genuine phenomenon—achieving near-GPT-4 performance at a fraction of the training cost. Whether V4 can create the same shockwave is a real question. Competitors have caught up significantly over the past year, and the market’s threshold for “impressive” keeps rising. But if CITIC Securities’ analysis holds—that V4 integrates the Engram module into its DSA+MoE architecture—it could deliver a qualitative leap in ultra-long context processing.
TSMC March Revenue Surges 45% YoY; AI Chip Demand Unaffected by Middle East Tensions
Source: https://www.taipeitimes.com/News/biz/archives/2026/04/11/2003855383
TSMC reported March revenue of NT$415.19 billion, up 45.2% year-over-year and 30.7% month-over-month. Q1 combined revenue broke the NT$1 trillion mark for the first time, reaching NT$1.134 trillion, up 35.1% YoY—the fourth consecutive quarterly record.
Analysts project TSMC’s gross margin could reach a historic high of 65%. AI chip demand continues to accelerate, unaffected by Middle East tensions.
My take: TSMC’s results are the most direct barometer of AI infrastructure investment热度. A 45% YoY single-month increase means tech giants are still accelerating AI chip spending. The Middle East conflict’s impact on the supply chain has been overestimated—high-end chip manufacturing and shipping routes don’t pass through the conflict zone. A projected 65% gross margin would be among the highest in semiconductor industry history, which also explains why Intel stock surged 24% this week.
Five Chinese Ministries Release “AI Human-Like Interaction Service Management Measures,” Effective July 15
Source: https://www.uuwatch.com/newsDetail?nid=1852
China’s Cyberspace Administration, along with four other ministries, jointly released regulations governing AI human-like interaction services. Key provisions include: banning virtual family member and companion services for minors; implementing a categorized, tiered regulatory approach with a “tolerant and prudent” framework.
The measures take effect on July 15, 2026.
My take: This is the world’s first dedicated regulatory framework for AI anthropomorphic interactions—China is once again ahead on AI governance. Banning virtual companion services for minors is reasonable—the psychological impact on adolescents hasn’t been adequately studied. But the specifics of “categorized tiered regulation” remain unclear: where do you draw the line on “anthropomorphic” (does a customer service chatbot count?). For the industry, clear rules are ultimately better than ambiguous gray areas—they enable long-term planning and investment.
WeChat Cracks Down on AI-Automated Content Creation; AI Writing Account Generating ¥2M Annually Banned
Source: https://www.uuwatch.com/newsDetail?nid=1852
WeChat Official Accounts recently added a “non-human automated content creation” rule, explicitly prohibiting the use of AI, scripts, APIs, or other automated methods to replace human content creation. A couple who generated公众号 articles through AI and sold AI creation platform services—claiming annual revenue of ¥2 million—had their account “爆了么 AI” banned.
My take: WeChat’s new rule hits at the core debate around AI content creation: how much “human authenticity” does a platform need? From the platform’s perspective, mass AI-generated content degrades information quality and harms user experience. From the creator’s side, AI is just a tool—like Photoshop was to photography. The key distinction may not be “whether AI is used” but “whether AI use is disclosed” and “whether AI-generated content is human-edited and curated.” Different platforms will likely develop different AI content policies going forward, and this will fundamentally shape the content ecosystem.