AI Daily Dev — Week 22

Pope Leo XIV Publishes the First Papal Encyclical on AI Alongside an Anthropic Co-Founder

Magnifica Humanitas released today at the Vatican Synod Hall; signed May 15, 135 years to the day after Leo XIII's Rerum Novarum on labor and industrialization.
Anthropic co-founder Chris Olah on the launch panel — first time a pope has personally attended the press conference for one of his own social encyclicals.
Argues AI's simulation of faces, voices, empathy, and wisdom 'encroaches on the deepest level of communication, that of human relationships' and explicitly rejects transhumanism.
Frames AI as the second industrial revolution and the next great test for Catholic social teaching on dignity, labor, and justice.

industry vaticannews.va

Project Glasswing's first month surfaced 10,000+ high/critical-severity findings across ~50 partner orgs; 1,094 confirmed critical so far.
Only 97 patched upstream — some open-source maintainers have asked Anthropic to slow the disclosure rate.
Standout finds: a 27-year-old OpenBSD kernel flaw and a 16-year-old FFmpeg bug; Cloudflare reported ~2,000 bugs (400 high/critical), Mozilla patched 271 in Firefox 150.
Anthropic committing $4M to OSSF Alpha-Omega and ~$100M in Mythos credits to maintainers; Mythos itself stays gated to the defensive consortium.

research anthropic.com

Second $30B round inside a single calendar year — Series G in February was $380B post-money, this one lands above $900B.
Co-led by Sequoia, Dragoneer, Altimeter, and Greenoaks at ~$2B each; Founders Fund and General Catalyst also participating.
Pushes Anthropic past OpenAI's $852B March valuation as the world's most valuable AI startup for the first time.
Backed by Q2 guidance of $10.9B revenue (more than double Q1's $4.8B) and a first-ever quarterly operating profit.

industry bloomberg.com

Project Glasswing's first monthly update: ~50 partners using Mythos Preview have surfaced 10,000+ high- or critical-severity bugs across the world's most systemic software.
Cloudflare alone found 2,000 bugs (400 high/critical) with a false-positive rate it calls better than its human testers; Mozilla patched 271 Firefox vulns — a 10× lift over an older Claude model.
Anthropic separately scanned 1,000 OSS projects and flagged 6,202 high/critical vulns out of 23,019 issues; the new bottleneck is no longer finding bugs but verifying and patching them.
Front-paged on HN; Mythos itself stays locked — Anthropic still says no safeguards are strong enough to ship it.

research anthropic.com

SWE-Bench Pro jumps from 64.3% to 69.2%, ahead of GPT-5.5 and Gemini 3.1 Pro on Anthropic's runs; agentic computer use 82.8% → 83.4%, knowledge-work Elo 1753 → 1890.
Anthropic says Opus 4.8 is ~4× less likely than 4.7 to let its own code flaws slip past unremarked, and is more willing to flag uncertainty mid-task.
Same $5/$25 per million token pricing as 4.7; fast mode is ~2.5× quicker, plus a new claude.ai 'effort' control.
Claude Code v2.1.154 ships Dynamic Workflows: Claude writes a JS orchestration script, runs up to 16 concurrent and 1,000 total subagents per run.
HN front page within hours; an 'Is Opus 4.8 broken?' thread on file-reading reliability is already trending alongside it.

models anthropic.com

More Perfect Union surfaced an April 30 all-hands recording one day before mass layoff emails went out.
'Model Capability Initiative' captured keystrokes, mouse clicks, and screenshots from Gmail, GChat, VSCode, and internal AI assistant Metamate.
Zuckerberg on tape: average Meta employee is 'significantly higher' IQ than contractor labor — better training data.
1,000+ Meta workers have signed an internal petition to kill the program; EU employees exempt by GDPR.
Meta has not confirmed the recording's authenticity; story trended on Reddit and X all weekend, with 'train your replacement' the dominant framing.

industry theregister.com

New 'Locked Use' mode lets developers send Codex tasks from their phone and have it click, type, and navigate macOS apps with the screen off.
Apple authorization plugin temporarily unlocks the Mac, with short-lived auth, covered displays, and instant relock on any local input.
Mac-only; blocked in the EEA, UK, and Switzerland; explicitly cannot drive Terminal, Codex itself, or system admin prompts.
Reaction split — power users see remote agent superpower, security folks compare it to giving a 'sketchy AI company HAL 9000 access' to your laptop.

tools macrumors.com

Early-stage deal would rent Azure capacity backed by Microsoft's own 3nm custom accelerator, launched in January.
Microsoft says Maia 200 delivers >30% better tokens-per-dollar than the previous silicon in its fleet.
Would make Anthropic the first major external Maia customer and add a fifth silicon vendor — alongside Nvidia, AWS Trainium, Google TPUs, and SpaceX/xAI Colossus.
Same week Anthropic locks in its $900B round; The Information and CNBC both confirm the talks.

industry cnbc.com

Yuxiang Lin's Claude Code plugin sits at 32K GitHub stars and added 5,604 in a single day — top trending AI repo this week.
Multi-agent pipeline scans a repo, builds an interactive graph of files, functions, classes, and dependencies, then opens a clickable web dashboard with plain-English explanations.
Also works with Codex, Cursor, Copilot, and Gemini CLI; ships a guided-tour mode for onboarding new engineers.
v2.7.3 released May 19; pitched explicitly at the 'opened a 50K-line repo and had no idea where to start' problem.

open-source github.com

Daniel Stenberg's 'The pressure' post: incoming security reports now run 4–5× higher than 2024 — more than one per day.
12 confirmed vulnerabilities with half the release cycle still left, a project record; on track for ~30 CVEs before mid-2026.
These are detailed, AI-assisted reports, not the old 'AI slop' — the crushing part is the sheer volume of high-priority work.
Stenberg says it's the first time his wife has raised concerns about his hours; picked up by LWN, Simon Willison, and HN.

community daniel.haxx.se