The AI That Hacks Governments — And the $30B Lab That Built It

🛡️ Project Glasswing: When AI Becomes the World's Best Hacker Anthropic launched Project Glasswing, a controlled initiative giving select organizations access to Claude Mythos Preview, its unreleased frontier model, to find and fix critical software vulnerabilities before malicious actors can exploit them. In just weeks of internal testing, Mythos Preview identified thousands of zero-day vulnerabilities across every major operating system and web browser, including a 27-year-old bug in OpenBSD. The implications are enormous: a model too dangerous to release publicly is being used to quietly patch the internet's most critical infrastructure. This is safety theater turned safety engineering — and it's rewriting how governments think about AI oversight. Crescendo AI Crescendo AI

💼 Anthropic Releases 10 Pre-Built Financial AI Agents Anthropic released ten preconfigured AI agents for the financial sector designed to automate typical tasks performed by investment banks, asset managers, and insurers. This isn't a demo — it's a direct bid for Wall Street's workflow stack. Packaged, vertical-specific agents are quickly becoming the enterprise go-to-market play for every major frontier lab. LLM Stats LLM Leaderboard

🤖 Meta Bets Big on Proprietary AI — and Bigger on Capex The proprietary model — a departure from Meta's open-source Llama strategy — delivers competitive performance on multimodal perception, reasoning, health, and agentic tasks at a fraction of the compute cost of its older Llama 4 mid-size variant. Meta simultaneously announced AI capital expenditures of $115–135 billion for 2026, nearly double last year's spending. For a company that built its AI reputation on openness, this pivot to closed models is a significant strategic reversal. Crescendo AI Crescendo AI

⚡ Google's TurboQuant Tackles the KV Cache Bottleneck Google's research team unveiled TurboQuant at ICLR 2026, an algorithm that significantly reduces the memory overhead caused by the KV cache, one of the biggest bottlenecks in running large AI models. Using a two-step process combining PolarQuant vector rotation and the Quantized Johnson-Lindenstrauss compression method, TurboQuant allows models with massive context windows to run far more efficiently. For engineers and infra teams, this is the kind of unglamorous-but-critical research that meaningfully expands what's deployable at scale. Crescendo AI Crescendo AI

This week crystallized what security researchers have been warning about for two years: frontier AI models are now capable of autonomous, end-to-end cyberattacks — and the gap between offense and defense is widening fast.

Two frontier models cleared a 32-step end-to-end cyber-attack range in a single month. Anthropic's Claude Mythos Preview did it first; GPT-5.5 followed three weeks later. The UK's AI Security Institute now estimates frontier cyber-offence capability is doubling every four months. Airstreet

The policy response is now catching up to the threat. The NSA and the Office of the National Cyber Director are expected to lead safety testing for these frontier models. While the government would not have a direct veto over releases, the framework grants early access to agencies to develop pre-release oversight procedures. This signals a structural shift: the "release with guardrails" era may be ending for the most capable models. devFlokers

Meanwhile, Mandiant's M-Trends 2026 report found that time-to-exploit has effectively gone negative — exploits are now routinely arriving before patches, with 28.3% of CVEs exploited within 24 hours of disclosure. AI is compressing attack timelines faster than defenders can respond. The arms race is real, and the scoreboard isn't even close right now. The Hacker News

The Hacker News | Air Street Press

The narrative around AI's commercial race just flipped. For the first time, Anthropic's annual recurring revenue eclipsed that of OpenAI. Anthropic reached an annualized run rate of $30 billion, while OpenAI trailed at $24 billion. devFlokers

This is more than a financial milestone. It reflects a major enterprise shift: Anthropic's safety-first positioning — once mocked as a commercial liability — has become its strongest sales argument with regulated industries (finance, healthcare, government). The company's Claude suite, its pre-built vertical agents, and Project Glasswing are all converging into a coherent enterprise stack.

Frontier labs became infrastructure companies. Anthropic took an additional $40B from Google and $5B from Amazon (packaged with $100B of AWS spend), and signed chip deals with Google and Broadcom reportedly worth hundreds of billions. The company is no longer just an AI lab — it's becoming critical computing infrastructure for the Western economy. Airstreet

Air Street Press | DevFlokers

Want to advertise?