AI News: April 15, 2026

1. MIT Technology Review: MCP Crosses 97M Installs as Frontier Models Reach Parity

Research. MIT Technology Review published a data-driven analysis of AI’s current state showing Gemini 3.1 Pro, GPT-5.4, and Claude Opus 4.6 within a few benchmark points of each other, making workflow quality a more meaningful differentiator than model selection. Anthropic’s MCP protocol has crossed 97 million installs with every major AI provider now shipping MCP-compatible tooling. The piece also notes that at current rates of progress, the best open-weight models are within 9 points of the top proprietary models on aggregate evals — a gap that continues to close. Source

2. Claude Mythos Evaluation Exposes Gaps in Europe’s AI Safety Infrastructure

Policy. The Decoder analyzed how Anthropic’s restricted access to Claude Mythos reveals a structural problem for European AI safety institutions: UK regulators independently evaluated the model’s cyberattack capabilities, while EU bodies had no access or visibility into the system. The article argues that Europe’s AI Act enforcement mechanisms were designed for known systems with public deployments, not gated research previews with sovereign-adjacent implications. The episode is prompting calls for a mandatory pre-release evaluation regime analogous to nuclear or biological dual-use controls. Source

3. Science Corp Preparing to Place First AI-Guided Neural Sensor in a Human Brain

Biotech. Science Corp, founded by former Neuralink co-founder Max Hodak, is preparing to implant its first neural sensor in a human patient. The device uses electrical stimulation patterns informed by AI models to encourage healing of damaged brain and spinal cord tissue, targeting neurological conditions rather than the motor control or communication use cases pursued by Neuralink. The approach positions AI as a treatment orchestration layer rather than a direct interface, with the first human trial expected within months. Source

4. Ukraine Captures Military Position Using Only Drones and Ground Robots

Defense. Ukraine reported the first confirmed capture of a military position accomplished entirely by unmanned systems — no human soldiers participated in the physical operation. A CSIS analysis accompanying the report examines how AI-assisted target identification, autonomous coordination between drone swarms and ground robots, and edge inference capabilities are reshaping tactical operations. The milestone marks a doctrinal shift from AI as a decision-support tool to AI as an operational executor in contested environments. Source

5. Vibe-Coding App “Anything” Rebuilds After Two App Store Removals

Developer Tools. AI coding app Anything was removed from Apple’s App Store twice — once for allowing users to generate code that Apple deemed policy-violating, and again after a modified version crossed similar thresholds. The company is now developing a desktop companion app outside the App Store to sidestep review constraints, while debating how much guardrails to impose on AI-generated output. The story highlights the emerging tension between AI coding tools and platform gatekeepers as vibe-coding apps generate increasingly capable code without human review steps. Source

6. Anthropic’s Momentum Is Prompting OpenAI Investors to Reassess Valuations

Market. A prominent investor with stakes in both Anthropic and OpenAI told TechCrunch that sustaining OpenAI’s current valuation requires assuming a $1.2 trillion IPO, while Anthropic’s $380 billion valuation looks more grounded given its enterprise revenue trajectory. The comment reflects a broader recalibration among institutional investors as Anthropic gains share among enterprise customers in coding and regulated industries. Anthropic has not announced IPO plans, but the appointment of Novartis CEO Vas Narasimhan to its board is being read as preparation for public markets. Source