LLMs are making formal software verification feasible at a much larger scale
What happened
Anthropic used its Claude model to formally prove security properties in the compiled RISC‑V code of 123 open‑source libraries, automatically generating assembly‑level proofs that the programs cannot crash or trigger memory‑safety errors under specified conditions. In the process Claude uncovered long‑standing bugs that had escaped detection despite extensive testing. The effort demonstrates that large language models can produce formal verification proofs at a scale previously limited to small, mission‑critical projects. This marks the first large‑scale application of AI to formal verification of production software at the assembly level. The work suggests that as model capabilities improve, formal verification could become a routine part of secure software development.
Why it matters
For ordinary readers, this means software we rely on could become more trustworthy as AI helps catch hidden bugs before they cause problems. It also hints that future updates and patches may be backed by mathematical guarantees rather than just testing.
Reality check
An automated, sceptical read of the sourceNotable. The use of an LLM to automatically generate formal proofs for over a hundred real‑world libraries is a significant step beyond earlier, smaller‑scale attempts, though it is still limited to specific assumptions and does not yet cover all bugs.
Is this a big deal?
Get stories like this every morning
The day's AI news in plain English, in your inbox at 8:30 IST. Free.
More from Anthropic
Anthropic Upgrades Claude Code Projects to Run Parallel AI Workers Hands-FreeAnthropic · 14 Nov 2026Claude designed FPGA scaler board for 6th-gen consolesAnthropic · 13 Nov 2026I asked Claude Fable 5.2 to turn my site into a 90s Silicon Graphics workstationAnthropic · 13 Nov 2026Anthropic is endorsing SB 53Anthropic · 13 Nov 2026Understanding and addressing AI harmsAnthropic · 13 Nov 2026This page was collected from anthropic.com and summarised automatically by AIPulsee's pipeline. See how it runs.