Brief IA

Apple and Anthropic: iOS 26.5 Secured with Claude

🛠️ AI Tools·Tom Levy·

Apple and Anthropic: iOS 26.5 Secured with Claude

Apple and Anthropic: iOS 26.5 Secured with Claude
Key Takeaways
1Apple has integrated Claude from Anthropic into iOS 26.5, fixing over 50 vulnerabilities.
2Two critical flaws, CVE-2026-28942 and CVE-2026-28952, were discovered thanks to Claude.
3Apple may delay the announcement of flaws found by Mythos, according to its disclosure policy.
💡Why it mattersThe collaboration between Apple and Anthropic marks a significant advancement in the use of AI for cybersecurity, potentially influencing other tech giants.
Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

📄
Full Analysis

Apple and Anthropic: iOS 26.5 Secured with Claude

In early April, Anthropic officially announced the launch of its Project Glasswing, an artificial intelligence cybersecurity initiative. Among the eleven launch partners are giants such as Apple, Google, Microsoft, NVIDIA, and JPMorgan. Five weeks after this announcement, Apple released the iOS 26.5 update, incorporating over 50 security patches. Two security vulnerabilities explicitly mention "Claude, Anthropic," marking a first for Apple.

Two Critical Vulnerabilities Identified

The first vulnerability, referenced as CVE-2026-28942, concerns WebKit. It allowed a malicious iframe to exploit the download settings of another site. Researchers Milad Nasr and Nicholas Carlini, members of Anthropic's red teaming team, were credited with this discovery, in collaboration with Claude. The second vulnerability, CVE-2026-28952, affected the system kernel, where an integer overflow could cause a system crash. This discovery is attributed to "Calif.io in collaboration with Claude and Anthropic Research."

It is important to note that Apple did not mention "Mythos" in its security notes. The term "Claude" can refer to any version of Anthropic's model. Neither of the two vulnerabilities was exploited at the time of their correction, classifying them as zero-day.

Comparison with Mozilla and Future Perspectives

The situation at Apple contrasts with that of Mozilla, whose Firefox 150, released on April 21, included 271 vulnerabilities discovered by Mythos Preview. Mozilla had access to the model since February, three months before the release, and CTO Bobby Holley documented the process.

Several reasons may explain this difference. Apple does not always credit the discovery method in its documentation, and Anthropic's responsible disclosure period extends to 90 days, potentially up to 135 days. Vulnerabilities identified by Mythos as early as April 7 may not be made public until July 2026, or even postponed until the release of iOS 27 for the less urgent ones. Apple may also want to minimize the exposure of its reliance on a third-party AI.

The full report of Project Glasswing is expected in July 2026. To date, VulnCheck has recorded 40 CVEs attributed to Anthropic among all publishers. The Bundesbank and the Swiss FINMA have been calling for European access to Mythos since late April, highlighting the importance of this technology beyond software security.

In the meantime, the iOS 26.5 update is highly recommended. It fixes fifty vulnerabilities, including 10 in WebKit, and is the most significant patch since iOS 26.2. All iPhone 11 and later models are affected, with patches also available for older models: iOS 18.7.9, iPadOS 18.7.9, iPadOS 17.7.11, iOS 16.7.16, iPadOS 16.7.16, iOS 15.8.8, and iPadOS 15.8.8.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.