Tuesday, 10 November 2026
AlphaProof Code formally verifies Signal's encryption; Fable 5.2 and Mythos 5.2 launch; AI agents formalize CPython and surface real bugs
Today's Lead
EngineeringGoogle DeepMind
Introducing AlphaProof Code: a breakthrough in verifying complex software at depth
Google DeepMind used AlphaProof Code, a multi-agent artificial intelligence (AI) framework, for formal verification. The team verified the compiled code of Signal's cryptographic protocol. Human experts chose the security goals. AI agents wrote the proofs. The agents produced over six million lines of Lean code. They flagged about one percent of the code for human review. The work took about four weeks, and expert teams usually need years.
Engineering
Anthropic
Introducing Claude Fable 5.2 and Claude Mythos 5.2
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, which use the same underlying model. Fable 5.2 is available to all users. Mythos 5.2 is available to vetted organizations through verification programs. Anthropic reports that the models outperform Claude Opus 5.5 on almost all tested capabilities. Output is up to 30 percent faster than the previous version. Mythos 5.2 finds more security vulnerabilities than Opus 5.5, including 10.2 percent more in an internal Common Vulnerabilities and Exposures (CVE) test. Anthropic says over 90 percent of changes merged into its core codebase are now written by Claude.
broyojo.com
Formalizing CPython
The authors used artificial intelligence (AI) agents to translate CPython, the reference Python interpreter, into Lean, a proof assistant. Each function is copied, tested on at least 50,000 inputs, proved, and reviewed. So far, 4,071 of 26,004 functions are accepted, and the project found 418 bug reports. About 55 of these can crash the interpreter or corrupt memory.
Martin Fowler
Long-horizon agents and durable execution
Long-horizon agents pursue goals over days or weeks, such as customer onboarding or sales deals. Most of their time, these tasks wait for people and external systems. The agent must survive process restarts, failed calls, and duplicated messages. The article proposes durable execution, which records each external step in a write-ahead log. After a crash, the program replays the log and reuses the recorded results. Durable execution alone does not make an agent correct.
evanxmerz.com
AI usage generates a fragile codebase
Evan X. Merz argues that heavy reliance on artificial intelligence (AI) tools that write code makes software fragile. The tools write shallow tests and split logic into extra helper methods. Code review cannot keep pace, and fewer developers understand the system. The author expects more incidents, slower recovery, and hidden errors. He concludes that quality products need human judgment, while cheap products may accept machine-made software.
Humanities
Aeon
The wholehearted life
Ben Daitz profiles Roshi Joan Halifax, a medical anthropologist and caregiver. Her work with people who are dying grew from hard personal experience. A crisis followed her research with lysergic acid diethylamide (LSD) on patients who were dying. Buddhist meditation gave her a steadiness that the drug could not. She was ordained as a Zen priest in 1996.