DeepMind's AI verified key security guarantees in Signal's protocol
AlphaProof Code checked parts of Signal's encryption protocol in four weeks and found no new flaws, WIRED reports; the full app was not covered.
Key takeaways
- Google DeepMind used an AI system called AlphaProof Code to formally verify a set of key cryptographic guarantees in the Signal protocol, WIRED reports.
- The four-week project found no new vulnerabilities in the parts it examined and flagged long-fixed flaws in older Signal versions.
- The analysis did not cover the whole Signal app, and Google did not disclose what it cost.
- Signal's CTO said the approach goes beyond earlier verification work that covered smaller portions of the code; DeepMind says scaling to all software is still far off.
- NextFish found no DeepMind post or paper on the work; the account relies mainly on WIRED's reporting.
Google DeepMind researchers used an AI system called AlphaProof Code to mathematically prove a set of core cryptographic guarantees in the Signal protocol, WIRED reported on Tuesday. The project took four weeks and covered parts of the protocol, not the full Signal app.
The account below rests mainly on WIRED's reporting and the statements it quoted. NextFish could not find a DeepMind blog post or paper describing the work at the time of writing.
What did DeepMind do?
Formal verification uses mathematics to show that software behaves as intended. According to WIRED, DeepMind carried out what the magazine describes as a proof-of-concept project using AlphaProof Code, a new AI framework, and checked a number of the protocol's key guarantees. WIRED does not detail exactly which properties or which parts of the code were covered. The Signal protocol also protects messages in Google Messages and WhatsApp, WIRED notes.
The researchers found no new vulnerabilities in the parts of the protocol they examined. They also ran the system on older Signal versions containing known flaws that were fixed long ago; it identified those problems and showed that those versions could not be formally verified, WIRED reported. WIRED reports that the team is publishing the specification and the proof it built, to help others working on formal verification; NextFish had not found where they were published at the time of writing.
What did Signal and DeepMind say?
Signal's chief technology officer, Ehren Kret, said in a statement that earlier verification work had covered smaller parts of the code and that "Google DeepMind's approach enables significant expansion". He added that the result gives Signal "even greater confidence in our code's privacy guarantees."
DeepMind research scientist Swarat Chaudhuri told WIRED there is "a long distance to go" before the method can be applied to all software. He said the team's process keeps humans reviewing the work at key points, rather than letting AI write both the goals and the proofs. Pushmeet Kohli, DeepMind's vice president of research, who is also chief scientist at Google Cloud, said banks and other companies that rely on such assurances and worry about bugs "could now say, 'I can prove these properties.'"
What are the limits?
The analysis was limited in scope and did not verify the entire Signal mobile app, WIRED reports. The researchers also say they had a head start, to a degree, from some formal verification work the Signal Foundation had already carried out.
Cost is an open question. DeepMind did the work free of charge as a research project, Google declined to say what it cost, and the company would not say whether it plans a commercial verification product or service. WIRED notes that it remains unclear whether AI-driven formal verification can scale to large amounts of code at a workable price.
Is anyone else doing this?
Signal is also part of Signal Shot, a public project with the Beneficial AI Foundation and the Lean FRO to verify the protocol and its actual Rust code, libsignal, using the Lean proof assistant. The author of an April blog post about the project wrote that he expects AI models to handle routine formalization while people pick the abstractions and look for subtle flaws. WIRED's report does not say whether DeepMind's work is connected to Signal Shot.
Questions readers are asking
What is formal verification?
It is a way of using mathematics to show that software behaves as intended. DeepMind used an AI system to produce such proofs for parts of the Signal protocol.
Did DeepMind find any problems in Signal?
According to WIRED, it found no new vulnerabilities in the parts it examined. When tested on older versions with known, long-fixed flaws, the system identified them.
Does this mean the Signal app is fully proven secure?
No. WIRED reports that the analysis was limited in scope and did not verify the entire Signal mobile app.
Will DeepMind sell this as a product?
Google declined to say whether it plans a commercial formal verification product or service, and did not share the cost of the analysis.
Sources
- #ai-security
- #formal-verification
- #signal
- #google-deepmind
- #cryptography
Get the day's top AI & tech stories on WhatsApp: Follow the NextFish channel
Get stories like this every morning.
Related
Cybersecurity4 min
ConnectWise urges ScreenConnect client updates as SCKit targets technicians
Microsoft, Huntress and Blackpoint have analysed the toolkit used against MSPs. It is built to jump to technicians, but researchers say they have not seen that work.
Cybersecurity3 min
Landmark Group says Shukran loyalty members' personal data was accessed
Names, contact details and dates of birth were involved, the Dubai retailer says, but not card details, passwords or points. It has not said how many members were affected.
Cybersecurity3 min
Operation Endgame takes down Gootloader, CastleRAT and StealC servers
The Justice Department says Gootloader infected at least 25,000 computers and was used against at least 44 US ransomware victims. Europol counts 235 servers hit in all.
Cybersecurity2 min
TSMC says leaked employee data matches records at a cloud provider
The chipmaker says its own systems show no sign of compromise after a seller offered a 6.9GB file of employee records on hacker forums for three bitcoin.