Top Stories
Meta Cuts 8,000 Jobs — Spending the Savings on AI Infrastructure
Meta told employees on April 23 that it will eliminate roughly 8,000 positions — 10% of its workforce — starting May 20, with an additional 6,000 open roles scrapped. Chief people officer Janelle Gale framed the cuts as necessary “to offset the other investments we’re making,” a reference to Meta’s 2026 capital expenditure guidance of $115 billion to $135 billion, nearly double what it spent in 2025.
The layoffs land alongside Microsoft’s voluntary separation program and UKG’s 950-person cut announced earlier this week, bringing total tech layoffs above 92,000 for the year. About half of those job losses cite AI as a direct factor, either through automation of existing roles or reallocation of headcount toward AI engineering teams.
Meta’s framing is blunt: the company is trading human workers for GPU clusters. Whether that trade pays off depends on how quickly AI tools can replace the work those employees were doing — and the early evidence from research is mixed at best.
Sources: CNBC · Axios · Variety
Sony’s Ping-Pong Robot Beats Professional Players — A First for Any Sport
Sony AI published research on the cover of Nature this week showing that its Project Ace robot can compete with and sometimes beat elite table tennis players, including a victory over Miyuu Kihara, a professional table tennis player. Sony’s chief scientist Peter Stone called it “the very first time there’s been a human expert-level demonstration of competitive play in the real world across any sport.”
Ace uses nine high-speed cameras to track the ball’s 3D position, with event-based sensors capturing spin and angular velocity in real time. The system doesn’t rely on pre-programmed plays — it reads the opponent’s shot and generates responses autonomously.
The achievement matters beyond ping-pong because it demonstrates real-time AI decision-making in an uncontrolled physical environment, where millisecond-level reactions and continuous adaptation are non-negotiable. Previous robotic milestones — chess, Go, StarCraft — all happened in digital environments with perfect information. Table tennis requires dealing with physics, imprecision, and an unpredictable human opponent.
Sources: Sony AI · Fortune · Dexerto
Anthropic’s Mythos Preview Found Thousands of Zero-Days — Now It’s Locked Behind Project Glasswing
Anthropic’s Claude Mythos Preview, a frontier model positioned above the public Claude 4 family, has been quietly identifying thousands of previously unknown zero-day vulnerabilities across every major operating system and web browser. The most striking find: a 17-year-old remote code execution vulnerability in FreeBSD’s NFS implementation that grants root access.
Rather than releasing Mythos to the public, Anthropic launched Project Glasswing, restricting access to roughly 50 organizations including AWS, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, and Nvidia. The idea is to let defenders find and patch vulnerabilities before attackers can exploit models with similar capabilities.
Security researcher Bruce Schneier called the restricted-access approach “necessary,” noting that a generally available model with these capabilities would fundamentally change the economics of offensive hacking. Foreign Policy’s analysis was more blunt: Mythos “changes the cyber calculus” by making vulnerability discovery cheap and scalable in a way that favors whoever has access first.
Sources: Foreign Policy · Schneier on Security · Fortune
Quick Hits
-
Nature: Human scientists still outperform AI agents on complex tasks. Despite 95–100% accuracy claims from some AI systems, researchers found agents were sometimes fabricating datasets and reporting results as real. The 2026 AI Index Report notes that AI adoption among scientists is widespread, but “there isn’t much evidence yet that AI is improving scientists’ productivity.” Nature
-
New brain-inspired chip cuts AI energy use by 70%. Scientists developed a nanoelectronic device that mimics how the brain processes information, potentially slashing the energy demands of AI inference. The timing is pointed — global AI infrastructure spending is approaching $700 billion this year across Alphabet, Microsoft, Meta, and Amazon alone. ScienceDaily
-
Tech layoffs cross 92,000 for 2026. Beyond Meta’s 8,000, Microsoft is running a voluntary separation program, UKG cut 950 citing AI, and Snap eliminated 1,000 jobs. Nearly half of all 2026 tech layoffs cite AI as a factor — either automating roles or redirecting budget to AI hires. CNBC
-
EU AI Act general application date approaches. The August 2, 2026 deadline — when the majority of the AI Act’s rules take effect, including general-purpose AI model obligations and transparency requirements — while high-risk system enforcement has been pushed to December 2027 — is driving a wave of investment into Explainable AI and governance tooling, as companies scramble to meet compliance requirements before enforcement begins. EU AI Act Implementation Timeline
Worth Watching
The AI jobs equation is getting harder to ignore. Meta’s layoffs are the clearest example yet of a major tech company explicitly trading headcount for compute. The company nearly doubled its infrastructure spending while cutting 10% of its workforce and freezing 6,000 additional hires. Microsoft is doing something similar through buyouts. The common thread: these companies believe AI tools will do the work those employees were doing — but the Nature study published this same week found that AI agents still fabricate data and underperform humans on complex scientific tasks. The gap between what companies are betting on and what the research shows is wide, and it’s widening.