Anthropic's Mythos Stays Under Wraps While April Becomes AI's Biggest Month

Anthropic's Mythos Stays Under Wraps While April Becomes AI's Biggest Month

Anthropic released Claude Mythos Preview yesterday, but you can't use it [4]. The cybersecurity-focused frontier model is locked to select partners like CrowdStrike through "Project Glasswing" — apparently it's too good at finding security flaws to release publicly [5]. Leaked reports suggest we're looking at 10T parameters, which would explain the caution.

April 2026 is shaping up as the most consequential month in AI model releases. We're expecting xAI's 6T parameter Grok-5, OpenAI's GPT-5.5 post-training update, and Huawei's 1T DeepSeek V4 all this month [4]. Researchers are calling it history-making, and for once, the hype might be justified.

Emad Mostaque is already questioning whether Mythos leaked "Claude Code" — whatever that means. But when Anthropic is too nervous to do a public release, you know we're entering new capability territory.

Eclipse Ventures Raises $1.3B for Physical AI and Defense

Eclipse Ventures closed $1.3B across two funds yesterday — $720M for early-stage, $591M for later-stage investments [6]. This brings their total assets under management to around $10B, with a focus on AI infrastructure, physical AI, robotics, manufacturing, defense, transportation, and energy [7].

These are the folks who backed Cerebras early, so they know how to spot AI infrastructure plays before they become obvious [8]. The fund size signals where smart money thinks AI is heading: out of the cloud and into the physical world. When VCs are writing checks this big for "physical AI," it's not speculation anymore — it's preparation for the next wave.

Harvey Legal AI Hits $11B Valuation

Legal AI startup Harvey raised $200M at an $11B valuation in March, with GIC and Sequoia co-leading [9]. That's up from $8B in December 2025 and $5B earlier that year — the kind of valuation hockey stick that makes other vertical AI companies pay attention [10].

Harvey is building AI agents specifically for law firms and enterprises, with embedded legal engineering teams [11]. This isn't just about replacing paralegals with chatbots — it's about rebuilding how legal work gets done from the ground up. When Sequoia "triples down" on a legal tech company, they're betting that industry-specific AI agents will eat entire professional service categories.

EU AI Act High-Risk Compliance Starts August 2nd

The EU AI Act's high-risk system obligations kick in August 2, 2026 [12]. If you're building credit scoring, fraud detection, or similar systems, you'll need risk management frameworks, assessments, human oversight, and conformity checks [13]. The good news? Systems deployed before August 2nd get grandfathered in, creating a brief window to avoid early compliance headaches [14].

EU member states need AI regulatory sandboxes ready by then too. Smart companies are already inventorying their AI systems and building governance frameworks now, rather than scrambling in summer.

Romania Pitches Itself as Europe's Low-Cost AI Hub

Romania is making a play to become Europe's AI development center, leveraging 142,000+ IT professionals, strong English proficiency, and costs 30-50% below Western Europe [15]. With elite engineering talent and full EU access, they're positioning as the obvious choice for AI R&D operations [16].

The pitch makes sense on paper — skilled developers, regulatory certainty, and significant cost savings [17]. Whether Romania can overcome brain drain and build the ecosystem depth that AI companies need remains to be seen, but they're making the right moves with AI-focused events and R&D investment.