19 novembre 2026
Lebanon’s citizen media
All sections

News

Signal: what AlphaProof Code really proves

AlphaProof Code combines human experts, artificial intelligence agents and mathematical verification. The work announced by DeepMind is about a precise implementation of the Signal protocol. Its interest is based on the demonstrated properties, but also on the assumptions and components that remain outside the verified perimeter.

pexels-photo-576739.jpeg
Photo by Jonas Svidras on <a href="https://www.pexels.com/photo/black-iphone-7-displaying-at-09-56-576739/" rel="nofollow">Pexels.com</a>
Beta translationThis article is an automated beta translation. Please use caution and verify sensitive details against the French original when needed.

Artificial intelligence that judges a software safe does not alone provide proof of security. AlphaProof Code is precisely trying to overcome this difference. Presented on November 10, 2026 by Google DeepMind, the device brings together human experts, agents from the AI and a mathematical auditor to examine the code of the Signal protocol. The interest of this advance lies as much in what it demonstrates as in the explicit boundaries of its demonstration.

AlphaProof Code: Evidence, not a confidence mark

In the operation described by DeepMind, humans first participate in defining the properties sought and validate the specifications. Officers then produce the evidence. The result therefore depends not only on the ability of a model to write a convincing explanation: it must go through a formal audit.

The language used, Lean, is used to represent theorems and their evidence in a form controlled by a separate software. Its documentation explains that evidence can be independently verified. This is the difference between asking a system to defend a conclusion and asking it to provide a reasoning whose stages satisfy explicit rules.

This separation of roles is central. The tool that offers evidence is not the one that is asked to believe on word. But logical control does not decide, instead of researchers, whether the initial question covers all the significant risks to a user.

The precise scope of the result on Signal

The work is about libsignal, the protocol reference implementation, compiled for an AArch64 Linux environment. Researchers announce about 6.5 million lines of Lean evidence. They are based on some 26,000 lines of specification and description of machine operation which must be considered reliable.

The demonstrations include the initial establishment of the connection with PQXDH and the Triple Ratchet, used to change the keys. They concern confidentiality, authentication and protection properties after certain compromises. Low-level cryptographic components remain represented by contracts and assumptions.

This result is therefore not a certification of all Signal applications, each phone or all possible configurations. To test their approach, the researchers also reintroduced two old defects already corrected: the modified versions violated the expected properties. This is not the announcement of two new flaws in the current application.

Why keys change during a conversation

The technical documentation of Signal allows to understand the issue. In the Double Ratchet mechanism, each message uses newly derived keys. The old keys must not be able to be found from the next ones. New cryptographic exchanges also feed the mechanism to protect future communications.

The aim is to limit the extent of damage when a key is compromised. The protection of past messages and the possibility of finding security for the next respond to two different problems. The Triple Ratchet combines mechanisms for hybrid security, incorporating a post-quantum component.

This architecture does not transform a persistent intrusion into an incident without consequence. The documentation distinguishes recovery after compromise from situations where the attacker retains access to secrets. The renewal of the keys must therefore be understood with the conditions under which it can fulfil its role.

« Post-quantum » does not mean without hypothesis

PQXDH intervenes at the beginning of the exchange. It allows two people to establish a shared secret, including when the recipient is offline but has deposited the necessary information on a server. The published version of its specification distinguishes a persistent post-quantum confidentiality and mutual authentication that still depends on a classical cryptographic hypothesis.

These distinctions avoid summarizing safety with a label. Evidence must specify what an opponent is supposed to be able to do, what functions are considered safe and what part of the system is being studied. The progress announced by DeepMind concerns the verification of properties on real code. It does not remove the choice of assumptions or the need to examine them.

Illustration photograph: smartphone, archive of 2017, Jonas Svidras / Pexels. It does not represent the search device.

Report an error

Include the passage and proposed correction. Avoid personal information. The newsroom keeps reports for 90 days. Privacy

· · · ·

Libnanews
Newsdesk Libnanews - translated by IA

Libnanews est un site d'informations en français sur le Liban né d'une initiative citoyenne et présent sur la toile depuis 2006. Notre site est un média citoyen basé à l’étranger, et formé uniquement de jeunes bénévoles de divers horizons politiques, œuvrant ensemble pour la promotion d’une information factuelle neutre, refusant tout financement d’un parti quelconque, pour préserver sa crédibilité dans le secteur de l’information.

All articles by this author →

Comments

Laisser un commentaire

Votre adresse e-mail ne sera pas publiée. Les champs obligatoires sont indiqués avec *