April 08, 2026 – AI Daily Recap

The AI world shifted dramatically today as Anthropic unveiled Claude Mythos Preview — a model it considers too powerful for public release — alongside a $100 million cybersecurity initiative, while Z.ai’s open-source GLM-5.1 claimed the top spot on SWE-Bench Pro and a 26-person startup proved you don’t need billions to build a frontier model.

Anthropic Launches Project Glasswing With Claude Mythos Preview

Anthropic made the biggest splash of the day with the announcement of Claude Mythos Preview, a general-purpose language model the company describes as strikingly capable at computer security tasks. Rather than releasing it broadly, Anthropic built Project Glasswing, a cybersecurity initiative that pairs the model with twelve major technology and finance companies — including Apple, Microsoft, Google, AWS, Nvidia, and JPMorganChase — to find and patch vulnerabilities across critical infrastructure. According to Anthropic, Mythos Preview has already identified thousands of high-severity zero-day vulnerabilities in every major operating system and web browser. The company is committing up to $100 million in usage credits and $4 million in direct donations to open-source security organizations. Anthropic also revealed its projected annual revenue has tripled in 2026 to more than $30 billion, and the company is reportedly evaluating an IPO as early as October.

Z.ai’s GLM-5.1 Claims Open-Source Crown on SWE-Bench Pro

Z.ai released GLM-5.1, a 754-billion-parameter Mixture-of-Experts model that has claimed the number one open-source position and third place globally on SWE-Bench Pro with a score of 58.4, beating models from both OpenAI and Anthropic on coding tasks. What sets GLM-5.1 apart is its focus on long-horizon autonomous tasks — the model is engineered to maintain goal alignment over execution traces spanning thousands of tool calls, and Z.ai demonstrated it running autonomously for eight hours to build a Linux desktop environment from scratch. The model uses 40 billion active parameters despite its massive total count, keeping inference costs manageable. Available under the MIT license on Hugging Face, GLM-5.1 comes from a company that listed on the Hong Kong Stock Exchange earlier this year with a market capitalization of $52.83 billion.

Anthropic Poaches Microsoft’s Eric Boyd to Lead Infrastructure

In a move that underscores its rapid growth trajectory, Anthropic hired Eric Boyd from Microsoft to head its infrastructure expansion. Boyd spent more than 18 years at Microsoft, where he oversaw the hardware and software needed to host both OpenAI and Anthropic models on Azure’s cloud platform. Originally joining Microsoft as a manager leading BingAds development, Boyd rose to become president of the AI Platform in 2015 and was later tapped by CEO Satya Nadella to lead the Azure AI team. With Anthropic’s revenue now exceeding $30 billion annually and demand surging, CTO Rahul Patil said Boyd’s enterprise-scale infrastructure experience will help the company meet record customer demand.

Arcee’s 26-Person Team Releases Frontier-Class Open-Source Model

In one of the more inspiring stories of the AI era, Arcee AI — a scrappy 26-person startup — released Trinity Large Thinking, a 400-billion-parameter open-source model that CEO Mark McQuade claims is the most capable open-weight model ever released by a non-Chinese company. Arcee trained all of its Trinity models in six months for a total of just $20 million, using 2,048 Nvidia Blackwell B300 GPUs — a fraction of the resources available to labs like Meta or Google. Released under the Apache 2.0 license, the model aims to give Western companies a truly open alternative to Chinese-developed models, with both on-premises deployment and cloud-hosted API options available.

Bain Cuts Ties With Megaspeed Amid Chip Export Investigation

Bain’s data center division has severed its relationship with Megaspeed, a firm currently under U.S. investigation for allegedly helping Chinese companies circumvent Nvidia’s AI chip export restrictions. Megaspeed — formerly 7Road International, a Chinese gaming company with ties to the state — rapidly became Nvidia’s largest buyer in Southeast Asia, importing at least $4.6 billion worth of hardware and acquiring more than 136,000 GPUs. U.S. officials and Singaporean authorities are examining whether Megaspeed acted as a conduit for restricted chips ultimately destined for China. The move comes as enforcement intensifies: earlier this month, federal prosecutors filed criminal charges against two former Supermicro logistics managers in a related chip smuggling case.

Amazon Bedrock Launches Projects API for AI Cost Management

Amazon Web Services introduced the Bedrock Projects API in its Mantle inference engine, addressing a growing pain point for enterprises deploying AI at scale: knowing where the money is actually going. The new feature lets organizations create individual projects to isolate workloads, assign IAM-based access control, and tag usage for detailed cost tracking across business units, applications, and environments through AWS Cost Explorer. With AI inference costs becoming a significant line item for many companies, the ability to attribute expenses to specific workloads rather than a single undifferentiated bill could help enterprises make smarter decisions about model selection and optimization.

The Big Picture

Today’s news paints a picture of an AI industry that is simultaneously accelerating and fragmenting. Anthropic’s decision to gate Claude Mythos behind a cybersecurity initiative rather than release it publicly signals that the most powerful models may increasingly be deployed through controlled channels rather than open APIs. Meanwhile, the open-source world is thriving from both ends of the scale — Z.ai’s $52-billion-market-cap effort and Arcee’s $20-million shoestring budget both produced genuinely competitive models released under permissive licenses. On the infrastructure side, talent is flowing toward the companies with the most momentum (Anthropic poaching from Microsoft), enforcement around chip export controls is tightening (the Megaspeed fallout), and the nuts-and-bolts work of making AI economically sustainable is getting more attention (Amazon’s cost management tools). The industry is growing up fast, and the stakes are rising to match.

Sources

• Anthropic debuts preview of powerful new AI model Mythos in new cybersecurity initiative — TechCrunch
• Anthropic says its most powerful AI cyber model is too dangerous to release publicly — VentureBeat
• Claude Mythos Preview System Card — Anthropic
• AI joins the 8-hour work day as GLM ships 5.1 open source LLM — VentureBeat
• GLM-5 on Hugging Face — Z.ai
• Anthropic Hires Microsoft’s Eric Boyd to Lead AI Infrastructure Expansion — Bloomberg
• Tech Moves: Microsoft leader jumps to Anthropic — GeekWire
• I can’t help rooting for tiny open source AI model maker Arcee — TechCrunch
• Nvidia’s biggest Southeast Asia customer exposes the limits of U.S. AI export controls — Tom’s Hardware
• Amazon Bedrock announces Projects API — AWS