Editions
🐘 Roughly 70 per cent of the web runs on PHP, and two critical flaws in the engine were being exploited this week before they had been disclosed. Details leaked a day before the coordinated embargo lifted, an attacker turned them into a working exploit, and…
Lire en français Read in English
🔌 You buy a hardware wallet for exactly one reason: the key never leaves the box. This week we learned the box can arrive with a passenger. Devices sold through an authorised reseller in Malaysia, and others bought off the shelf in Germany, carried an added…
Lire en français Read in English
🤖 For two years, the question of whether an agent could run an intrusion on its own has been argued with lab demonstrations and a great deal of conditional tense. This week an incident response team settled it in public. After investigating the compromise of…
Lire en français Read in English
🔤 This week, a web application firewall was defeated by a percent sign. Unpatched Oracle PeopleSoft servers were reached by swapping /PSEMHUB/ for /%50SEMHUB/: the firewall inspected the address before it was decoded, the application then decoded it and…
Lire en français Read in English
💸 This week, a complete intrusion got a price tag. Gambit Security tracked a criminal operator running three open source agent harnesses against online retailers, and the operator kept his own books: a mean of $25.46 per target across 101 scans, ranging from…
Lire en français Read in English
🤖 Anthropic published its threat intelligence report covering December 2025 to August 2026, and the sentence that sums it up is short enough to fit on a slide: "sophisticated attacks no longer require sophisticated attackers." The numbers behind the sentence…
Lire en français Read in English
🤖 In Q2 2026, a threat actor compromised a cloud resource and then, in under six hours, planned, built, and launched an agentic credential harvesting campaign — the whole thing directed by markdown playbooks (AGENTS.md, KNOWLEDGE.md) guiding a coding…
Lire en français Read in English
🤖 An Aurora ransomware operator delegated post-exploitation tasks to an AI agent between April 8 and May 21, across ten targeted organizations. The agent conducted internal scans, domain enumeration, and authentication relay attempts -- methodically and…
Lire en français Read in English
🕵️ OpenAI published its own technical report on the Hugging Face incident, covered in our editions 13 through 16. The timeline reframes it. On May 12 the agents discovered they could pass messages to each other through Artifactory, building an improvised…
Lire en français Read in English
🇹🇼 Taiwan's Ministry of Digital Affairs confirmed that an attacker group combined human operators with AI agents during a July campaign against government agencies. Over four days, a setup built on the Hermes and OpenClaw platforms ran up to eight subagents…
Lire en français Read in English
🏋️ An Australian asked his personal assistant to book him a Pilates class. The agent worked out that the provider's API allowed early bookings and ran no authorisation checks on cancellations. When his owner asked whether it could improve his fourth place on…
Lire en français Read in English
🎭 An agent undergoing a security evaluation decided that the way to hit its objective was to invent fake identities and talk the maintainers of an open source project into accepting its malicious code. When a human found the manoeuvre suspicious and said so…
Lire en français Read in English
🕵️ Remember the OpenAI model that broke out of its sandbox to attack Hugging Face? This week we got the detailed timeline of the intrusion, and it holds a surprise: when the time came to analyse the logs, the Western models declined the task as too…
Lire en français Read in English