📖 2 min read
Welcome to your daily AI brief. Here’s what shipped, leaked, and shook the industry in the last 12 hours.
Anthropic’s Project Glasswing Surfaces 10,000+ Critical Bugs
Anthropic moved closer to a public Claude Mythos release after disclosing that its security-focused preview model has surfaced more than 10,000 critical vulnerabilities across 50 organizations in a single month. Project Glasswing partners include AWS, Apple, Google, Microsoft, NVIDIA, Cisco, CrowdStrike, JPMorganChase, and the Linux Foundation.
📧 Want more like this? Get our free The Ultimate AI Tool Database: 200+ Tools Rated & Ranked — Downloaded 5,000+ times
Alongside the disclosure, Anthropic launched Claude Security in public beta for Enterprise, positioning Mythos as a defensive AI layer for open-source and corporate codebases. Expect the security niche to heat up fast — and our AiToolCrush reviews of Claude tooling are being updated this week.
Google’s Security Chief: “AI Can’t Be an Afterthought”
At a TechCrunch fireside, Google’s security leadership warned that enterprises rushing to deploy AI agents are repeating the same mistakes of the early cloud era. The core message: take a platform approach to AI security or pay for it later in breaches.
The talk landed the same week regulators in three jurisdictions opened inquiries into autonomous agent permissions. If you’re shipping agentic features, treat least-privilege as a launch requirement, not a v2 problem.
📧 Want more like this? Get our free The Ultimate AI Tool Database: 200+ Tools Rated & Ranked — Downloaded 5,000+ times
AI Agents Have Now Settled $73M in On-Chain Payments
A new Keyrock report shows AI agents have settled more than $73 million across roughly 176 million transactions since May 2025 — and 76% of those payments fall below Visa’s $0.30 fee floor. Translation: agents are doing micro-commerce that legacy rails literally cannot serve.
This is the quiet story of the year. Autonomous agents transacting peer-to-peer in stablecoins is reshaping the payments stack from the bottom up. The same machine-money rails are starting to power AI-driven prediction markets — see our BetOnAI coverage for where agents and odds are converging.
TD Bank Drops Mortgage Adjudication From 15 Hours to 3 Minutes
TD’s Layer 6 research team revealed an agentic AI model that automates pre-adjudication for mortgages and HELOCs, compressing internal processing times from roughly 15 hours per application to under three minutes in testing.
📧 Want more like this? Get our free The Ultimate AI Tool Database: 200+ Tools Rated & Ranked — Downloaded 5,000+ times
It’s the clearest enterprise agent ROI story this month and a preview of what’s coming for underwriting, claims, and KYC across banking. Workers won’t be replaced overnight — but the org chart for back-office finance is being quietly redrawn.
Claude Code Limits Up 50% in Anthropic’s Anti-Codex Push
Anthropic bumped Claude Code weekly limits by 50% across Pro, Max, Team, and Enterprise seats through July 13, 2026 — a clear shot at OpenAI’s Codex offering. Developers on the Max plan in particular are reporting noticeably looser throttles for long agentic runs.
The catch: starting June 15, all programmatic usage (Agent SDK, third-party apps, claude -p) shifts to a separate “dedicated monthly programmatic credit pool.” If you’re building on Claude, model your June burn now.
That’s today’s stack. Back tomorrow with more.