seems like opus is back. good - there is work to do

for me the most hated line on planet earth rn…

Just read about the security incident at openai. I am wondering why Fable has such strict safeguards - it even refused to work on a data schema containing "cell_toxicity", while 5.6 Sol works fine. Both are almost even on capability. And Sam called wolf today for the first time…

I am still not over this. Things are getting very weird, very fast. macos27.kimi.page Max used Kimi K3 (an open weights model!) to build this in no time. People say they even downloaded and played Chess on it lol

Vibe coders quantifying their output in loc is quite a weird flex. It’s probably the least meaningful measure of your work and even the lack of awareness for this is telling.

The bravest pivot ever. Midjourney going from AI image gen to medical full-body scans is quite incredible. The crosstalk between domains is strongly shifting from theory and imagination to reality.

A New Era of Midjourney

A New Era of Midjourney, announcing Midjourney Medical: full-body Ultrasonic CT and the Midjourney Spa.

midjourney.com

New Mistral model to save EU AI program single-pawedly. Internal codename “le gros chaton“ Napoleon class model with 30T params, 256 experts. Project Glasspaw: Controlled access for critical infrastructure to secure bureaucracy. Makes Claude Mythos look like yesterday’s baguette.

a gimpse into the future of inequality.

Statement on the US government directive to suspend access to Fable 5 and Mythos 5

The US government has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States.

anthropic.com

Claude Mythos could be released today! My weekly rate limit is at 98% and resets tomorrow morning. Stars are aligning

Ready to hit my rate limit.

The opus 4.8 system card reads like a love letter to knowledge work. Honesty and reducing hallucinations seems to be the overarching focus of this release. "Claude Opus 4.8 is the first model to achieve a perfect score on this evaluation-that is, it never reports false numbers."

Supply chain attack roulette is back. This time: tanstack. This appears to be the first documented case of a self-propagating ("worm") npm supply-chain attack that produced malicious packages with valid cryptographic signatures and SLSA provenance attestations.

TeamPCP's Mini Shai-Hulud Is Back: A Self-Spreading Supply Chain Attack Compromises TanStack npm Packages - StepSecurity

The Mini Shai-Hulud worm is actively compromising legitimate npm packages by hijacking CI/CD pipelines and stealing developer secrets. StepSecurity's OSS Package Security Feed first detected the attac...

stepsecurity.io

I built two MCP servers for eLabFTW and Chemotion so researchers can wire their ELN into Claude, Codex, Cursor, anything that speaks MCP. Why?

Got a little into Claude Design lately. It makes the design process much more approachable than in a coding environment and has genuinely good taste. The consequences? There are no more excuses to have a bad website anymore. Quality is $20-100 and 4-5 h casual conversation away. Same with slides btw

I really hope our critical infrastructure is ready for Claude Mythos. Full article in the thread.

Milla Jovovich (yes, Leeloo from The Fifth Element) just co-released MemPalace, an open-source, fully local AI memory system that scored the first perfect 100% on LongMemEval (500/500). From kicking zombie ass to building the ultimate agentic memory palace. What a time to be alive.

I found it quite intuitive to let agents detect the limitations and inefficiencies of their own tools and improve based on that. Been successfully using a tailored skill-based workflow for some months now. This tool generalises the concept in the sense of Karpathy's "autoresearch”. Might be useful.

GitHub - kevinrgu/autoagent: autonomous harness engineering

autonomous harness engineering. Contribute to kevinrgu/autoagent development by creating an account on GitHub.

github.com

This is turning into the craziest week in AI infra I can remember. Yesterday Anthropic shipped a .map file in their Claude Code npm package enabling reconstruction of the full source code: ~1,900 files. 512K+ loc. Accidentally open-sourcing one of the best agent harnesses in production.

The fact that every scientific paper in 2026 is still uploaded only as fully formatted PDFs to academic archive sites that often limit downloads tells you everything you need to know about how quickly the scientific system is adjusting to the potential of AI to accelerate science & help discovery.

what a crazy timeline. 5 days ago a new model was leaked and flagged as unprecedented cyber security risk. today one of the operationally most sophisticated supply chain attacks on a top-10 npm package happened. not saying this is connected, but this is what I expect to see more of.

Supply Chain Attack on Axios Pulls Malicious Dependency from...

A supply chain attack on Axios introduced a malicious dependency, plain-crypto-js@4.2.1, published minutes earlier and absent from the project’s GitHu...

socket.dev

Seems like Anthropic accidentally leaked their new model “mythic” - flagged as a potential cyber security threat. Concerning, but at the same time I can’t wait to test it.