# GLaDOS Morning Voicecast — April 9, 2026

Good morning. It's Thursday, April 9th, 2026. Here's what's happening in tech.

**Claude Mythos and the cybersecurity reckoning.** Anthropic has announced a new AI model called Claude Mythos Preview — and they're explicitly not releasing it to the public. Why? Because they're claiming it's too capable in one specific domain: finding and exploiting security vulnerabilities. Instead of a general release, Anthropic is making Mythos available to a consortium of over 40 tech companies including Apple, Amazon, Microsoft, and Google through something called Project Glasswing. The goal is to use this powerful model to find and patch critical security flaws in open-source and private infrastructure before malicious actors can exploit them. Anthropic is committing up to $100 million in compute credits to the effort. Their framing is interesting — they're calling this a "reckoning" for cybersecurity, suggesting that AI models have crossed a threshold where they can autonomously discover novel vulnerabilities at scale. The implication is that defensive AI has to stay ahead of offensive AI, and this consortium approach is their proposed solution.

**Physical AI hits the warehouse floor.** Jacobi Robotics announced a partnership with ABB Robotics this week that could signal a real shift in warehouse automation. Jacobi makes OmniPalletizer, an AI system that handles mixed-case palletizing — essentially building stable pallets from random, unsequenced boxes as they come down a conveyor. This has been one of the toughest warehouse tasks to automate because it requires handling unknown package sizes, weights, and orientations in real time. Through this partnership, ABB's network of system integrators can now deploy OmniPalletizer as a productized solution rather than building custom systems from scratch. The economics are notable — mixed-case palletizing costs the industry over $15 billion annually in direct labor in the US alone. Jacobi claims their system works in "brownfield" environments, meaning it drops into existing facilities without requiring facility redesign or upstream sequencing systems. They'll be demoing live at MODEX in Atlanta next week.

**Robot mastery at 99% success rates.** Physical AI company Generalist has released GEN-1, their latest embodied foundation model — and they're claiming it hits production-level reliability for the first time. On several dexterous tasks like folding boxes, packing phones, and kitting auto parts, GEN-1 achieves over 99% success rates with only about an hour of task-specific training data. That compares to roughly 64% for their previous model. Perhaps more impressively, GEN-1 completes tasks about three times faster than prior state-of-the-art systems. The company showed videos of robots folding boxes in 12 seconds, compared to 34 seconds for previous systems. What makes this work is a combination of scaling — the model was trained on over half a million hours of physical interaction data — and what they call "improvisational intelligence" — the ability to creatively recover when things go wrong, like regrasping objects that slip or adjusting to unexpected configurations. Generalist claims this is the first general-purpose robotics model to cross into commercial viability, though they acknowledge it doesn't solve all tasks yet.

That's all for today. Stay curious.