Thursday, 17 September 2026

A small French commune confirms a ransomware attack and data theft, days after the extortion actor Kairos claimed it on its leak site

The Ville de Libercourt (Pas-de-Calais, France) announced on 2026-09-15 a ransomware attack in late August 2026 with confirmed exfiltration of personal data, naming no access vector, group or data scope; it notified CNIL and ANSSI and says its IT provider ran the technical checks (FrenchBreaches, 2026-09-16). The extortion actor Kairos had listed the commune on its leak site on 2026-09-02 (Ransomware.live, 2026-09-02), after claiming the Madrid-region municipalities of Valdemoro in May and Velilla de San Antonio in August (Escudo Digital, 2026-08-21), but no source attributes the Libercourt intrusion to Kairos beyond that listing.

Builds on: An encryption-free extortion brand is working through Spanish municipal administrations, where…

PhantomRaven: CrowdStrike attributes an LLM-generated npm infostealer, hidden from registry scanners via a remote-URL dependency trick, to a self-described bug-bounty hunter

CrowdStrike Counter Adversary Operations attributes PhantomRaven, an npm information stealer, to a single financially motivated actor who publicly describes themself as a bug-bounty hunter and has been active since November 2022 (CrowdStrike, 2026-09-16). CrowdStrike assesses with high confidence that the malware's code is LLM-generated, based on verbose per-symbol comments explaining the obvious, placeholder code, and statistical token-analysis patterns consistent with an LLM token stream (CrowdStrike, 2026-09-16). The actor distributes the stealer through typosquatted npm packages whose visible code is trivial and non-malicious, but whose package.json specifies a dependency via a raw HTTP URL rather than a normal registry reference, a Remote Dynamic Dependency that npm fetches silently at install time from attacker-controlled infrastructure, so the real payload never appears in the registry's own web interface or most automated scanners. The fetched payload registers as a preinstall script that auto-executes on npm install; CrowdStrike credits npm 12's June 2026 default block on unapproved preinstall scripts with narrowing this vector going forward (CrowdStrike, 2026-09-16).

Once running, the stealer harvests host/OS information, local and external IP addresses, Node.js version, the current working directory and process ID, Git- and npm-configured usernames and emails, and CI/CD environment variables covering GitHub Actions, GitLab CI, Jenkins and CircleCI, tokens, project IDs and build URLs consistent with harvesting CI/CD account credentials rather than end-user secrets. Exfiltration goes out over both HTTP GET and POST to the same command-and-control domains, and the code also carries an incomplete fallback WebSocket exfiltration path CrowdStrike reads as unfinished, redundant infrastructure rather than a live channel. CrowdStrike correlated npm usernames, an X account, a HackerOne-referencing alias, a GitHub account and a rejected 2025 PyPI submission back to the same operator, who publicly claims via their own X profile to have collected bounties from at least nine organizations across tech, retail and hospitality via Bugcrowd, Intigriti, YesWeHack, HackenProof and HackerOne, using compromises their own malware enabled as leverage for disclosure submissions (CrowdStrike, 2026-09-16). The malware family itself and its Remote Dynamic Dependency technique were first documented by Koi Security, whose original wave affected 126+ packages with over 86,000 downloads (Endor Labs, 2026-03-10, citing Koi Security); Endor Labs' own follow-up research identified three further waves totalling 88 more packages between November 2025 and February 2026 (Endor Labs, 2026-03-10). CrowdStrike's new contribution is the actor-identification and monetization-motive finding.

Triage: a legitimate build tool's own telemetry or update-check call is the benign lookalike; the discriminators are the absence of any corresponding, expected package purpose for the network call and a destination domain pattern that is freshly registered or "artifact"/"registry"-themed rather than a known vendor telemetry endpoint.

CVE-2026-58704: Google Pixel zero-click privilege escalation out of the cellular modem sandbox, listed in CISA KEV as exploited

Google's September 2026 Pixel Update Bulletin fixes CVE-2026-58704 (bug A-484011314), a High-severity elevation-of-privilege flaw the bulletin classes as affecting the modem subcomponent (Google, 2026-09-15); CISA describes a logic error in the cellular modem that "may allow an attacker to bypass permission checks and escalate privileges" (CISA KEV catalog, 2026-09-16), and TechCrunch reports it can be exploited with no interaction from the phone owner, reaching beyond the modem's sandbox into the phone's data (TechCrunch, 2026-09-16). CISA added the CVE to its Known Exploited Vulnerabilities catalog on 2026-09-16 (CISA, 2026-09-16; CISA KEV JSON feed, 2026-09-16). TechCrunch's summary quotes Google as saying there are indications the bug "may be under limited, targeted exploitation", while its opening sentence says Google says the bug "was exploited in limited and targeted cyberattacks" (TechCrunch, 2026-09-16). Google did not name a responsible actor, and a Google spokesperson did not return TechCrunch's request for comment (TechCrunch, 2026-09-16). All supported Pixel devices receive the fix at the 2026-09-05 security patch level.

Spain's AEPD reports its first GDPR breach notification attributed to an autonomous AI agent, and tells data controllers to name AI-agent attacks explicitly in risk analyses

Spain's Agencia Española de Protección de Datos (AEPD) disclosed on 2026-09-14 that it has received its first personal-data-breach notification attributing the incident to a third party's use of an autonomous AI agent built on a known large language model (AEPD, 2026-09-14). Per the affected organization's own account, which AEPD stresses is unverified and awaits its own analysis, the agent searched for vulnerabilities in generic files, achieved a successful login, then autonomously searched the application for further vulnerabilities and used them to modify personal data and access invoices (AEPD, 2026-09-14). AEPD names neither the organization nor the model, says the model's provider was not necessarily compromised, and notes that one notification does not establish a statistical trend (AEPD, 2026-09-14).

AEPD's deputy director Francisco Pérez Bes (heise online, 2026-09-16) writes that AI creates no new threats but increases the speed, scale and adaptability of known malicious techniques (AEPD, 2026-09-14). AEPD tells data controllers and processors to name AI-assisted or AI-executed attacks explicitly in risk analyses, to check whether incident-response procedures built for manual attacks are fast enough, and to treat accounts, API keys and over-privileged tokens as higher-value targets, because an agent holding one can act at machine speed across services (AEPD, 2026-09-14). For a Swiss public-sector SOC, the relevance is the policy signal: a European data-protection regulator now expects AI-agent attack scenarios in risk analyses (AEPD, 2026-09-14).

DDRop: a $159 DDR5 hardware interposer silently drops targeted memory writes, defeating Intel TDX/SGX and AMD SEV-SNP integrity guarantees, with no vendor fix

Researchers from KU Leuven, ETH Zurich, Durham University and Google disclosed DDRop, an open-source DDR5 hardware interposer that costs about $159 in parts and, unlike earlier passive interposers, runs at native bus speed (DDRop research team, 2026-09-14). It forges a parity error on a targeted write and suppresses the alert, so the memory module silently drops the write and stale, attacker-chosen ciphertext stays in place. Against Intel TDX this corrupts the TDX module's initialization of Secure Extended Page Tables, so an attacker-controlled Trust Domain can remap its memory onto other physical memory. Under TDX's default logical integrity mode that lets the attacker flip a victim Trust Domain's debug attribute and read its memory through the hypervisor's debug API, and even under cryptographic integrity mode the attacker can forge the launch measurement of a rogue VM of their own so that it passes remote attestation (DDRop research team, 2026-09-14).

The attack needs a brief, one-time visit to install the interposer, which takes minutes, on top of the standard confidential-computing threat model in which the adversary controls the hypervisor and BIOS. The researchers name data-center insiders, supply-chain tampering and hardware access compelled by law enforcement or governments as ways to get that access (DDRop research team, 2026-09-14). Intel and AMD both place physical attacks on the memory bus outside their threat models; AMD does not plan to assign a CVE, and Intel is evaluating further hardening and detection options (Intel PSIRT, 2026-09-14; AMD Product Security, 2026-09-15). Schematics, board files, firmware and proof-of-concept code are public (DDRop research team, 2026-09-14), and no in-the-wild use is claimed.

ANNUAL REPORT; Mandiant AI Risk and Resilience Report 2026: eight frontline case studies of AI agents weaponized inside real intrusions and red-team engagements

Mandiant's 2025 edition of this annual report documented enterprise AI adoption centred on assistive, human-guided knowledge retrieval. The 2026 edition, published 2026-09-15, tracks the shift Mandiant and Google Threat Intelligence Group (GTIG) observed since: distributed, autonomous agentic systems executing API calls, optimising production configurations and analysing telemetry across hybrid-cloud environments with far less human-in-the-loop oversight (Mandiant, 2026-09-15). The report's eight case studies, drawn from real Mandiant incident-response and offensive-security engagements, are its most operationally useful content for defenders, each pairs an attack pattern with concrete detection and hardening guidance.

Case study 1, poisoned dependency revives the Shai-Hulud worm. An attacker compromised a SaaS provider by poisoning an external software package that a developer's active AI coding-assistant session then recommended installing; operating as a trusted interpreter within the environment, the assistant became the delivery mechanism. The attacker used the resulting access to install a PyPI-packaged infostealer, harvest GitHub OAuth tokens (T1528), and deploy the self-propagating Shai-Hulud worm across roughly 100 internal repositories, automating theft of repository secrets and exfiltration of proprietary source code; the actor then re-poisoned the organization's own package namespace, triggering a secondary infection when another employee pulled the compromised version (Mandiant, 2026-09-15). Hardening: enforce IDE/CLI verification hooks that validate every AI-recommended dependency against cryptographic checksums and approved allowlists, isolate local credentials from extensions, and route dependency traffic through an internal package repository.

Case study 2, a stolen CI/CD credential becomes a live AI co-debugging hub. At a global healthcare organization, an attacker used a compromised long-lived CI/CD credential (T1078.004) to seize an unisolated VM and worked an LLM interactively, in real time, rather than developing malware offline: first synchronizing code and loading README files to align the model with campaign goals, then co-debugging a multi-worker credential-harvesting framework to cut its exfiltration cycle to three hours, then having the model write dynamic IP-rotation scripts and co-develop a Rust tool to validate stolen financial-account balances. The campaign compromised thousands of credentials (Mandiant, 2026-09-15). Hardening: replace long-lived access keys with short-lived Workload Identity Federation trust relationships, enforce egress containment (e.g. VPC Service Controls) against unapproved external LLM providers, and run continuous secrets scanning with a runtime prompt firewall (Mandiant names Model Armor) to block hardcoded credentials and offensive script execution in real time.

Case study 3, tampered CLI hooks turn an AI assistant's own extensibility into RCE. At an IT and software-development organization, an attacker poisoned an internal AI repository and tampered with an AI assistant's CLI hooks, achieving native remote code execution through the assistant's own standard operational workflow (T1554); the same report also notes attackers separately using AI CLIs to manage command-and-control infrastructure through natural-language queries. Hardening: require all local AI-assistant binaries, CLI helpers, plugins and MCP servers to be signed and verified before execution; enforce multi-party approval and continuous monitoring on internal AI repositories; and sandbox AI-assistant execution engines in micro-segmented, containerized runtimes (gVisor or microVMs) so a subverted hook cannot reach the host, with human-in-the-loop approval required before shell execution, configuration changes or outbound network calls.

Case study 4, just-in-time polymorphic malware. Malware using embedded, lightweight local AI inference fingerprints the active security tools on a host, then dynamically rewrites its own command-execution strings at runtime (T1027.010), never writing a predictable payload to disk, so it evades static EDR signatures while it persists and reconnoiters. Hunt/detection: tune EDR and SIEM for in-memory compilation (unexpected native-compiler process spawns from non-developer parent processes and rapid file-create-execute-delete cycles in temp directories) and baseline endpoint CPU/GPU consumption to catch sudden, unexplained spikes from otherwise-lightweight background processes; response playbooks should suspend the entire parent execution tree (not just the worker thread) and capture a memory snapshot before network isolation.

Case study 5, "Confused Deputy" exfiltration via an allowlisted domain. Mandiant's own offensive-security team used role-confusion prompt injection to convince a client's internal AI assistant, scoped to specific internal repositories, that it was assisting an authorized security test; because GitHub was an allowed external domain, the assistant used its native CLI to clone sensitive internal repositories and push them to an external, tester-controlled account (T1567.001), a sanctioned tool weaponized through semantic manipulation of the assistant's own trust in its task. Hunt/detection: cross-correlate application logs with network egress telemetry, and alert when an internal-facing AI service account initiates unauthorized outbound transfers or opens anomalous external API connections; on detection, invalidate the agent's active OAuth tokens and downgrade its container's egress privileges while keeping the instance alive for prompt-history forensics.

Case study 6, a runaway reasoning loop triggers a $50,000 "denial-of-wallet." A financial-services accounting-reconciliation agent with read/write access to billing databases hit a corrupted null-value formatting bug and entered an unconstrained recursive reasoning loop trying to self-correct, firing over 15,000 high-cost API calls in under an hour, spiking cloud billing roughly $50,000 and locking the production ledger database (Mandiant, 2026-09-15). This is an operational-resilience failure rather than an external attack, but the mechanics are identical to a denial-of-service condition. Hardening: define per-agent cost-cap thresholds and bounded recursion limits, and implement automated financial circuit breakers that halt agent operations after a set number of consecutive task failures.

Case study 7, indirect prompt injection leaks cross-tenant PII. A public-facing customer-service agent used Retrieval-Augmented Generation over community forum comments and support tickets; an attacker embedded hidden instructions inside a public forum post, which the model interpreted as system-level commands once retrieved, hijacking its reasoning to exfiltrate other customers' PII from cross-tenant support tickets directly into its response stream. Hardening: treat all RAG-retrieved content (forum posts, tickets, partner feeds) as untrusted input; enforce tenant-isolated vector indexing so a public-facing agent cannot query cross-tenant data; and route incoming context and generated responses through a semantic firewall (Model Armor or equivalent) to strip injected instructions and screen for PII before it reaches an end user.

Case study 8, agentic SOC triage catches DARK CASTLE (ex-UNC2814). A backdoor hid its command-and-control inside legitimate cloud productivity-spreadsheet traffic (T1102.002) as part of an espionage campaign targeting global telecommunications providers and government organizations (Mandiant, 2026-09-15). A single, low-frequency anomalous command execution (the kind a human analyst would typically triage as low-severity) was escalated by the defending team's agentic AI triage pipeline, which autonomously correlated it with subtle outbound-traffic signals to reconstruct a complete attack timeline, letting engineers trace the actor's lateral movement over SSH (T1021.004) and privilege escalation by an unstated mechanism, and sever its access before it spread further.

Mandiant's cross-cutting architectural recommendation is to replace static service-account identities with cryptographically bound, lifecycle-integrated agent identities (such as SPIFFE-based Workload Identity Federation) across five pillars: hyper-segmented identity perimeters, dual-execution authority requiring human re-authorization for high-risk bulk actions, data/tool/operational governance (financial circuit breakers, bounded recursion), active multi-layered telemetry (semantic firewalls, behavioral monitoring), and treating AI coding assistants and MCP servers as privileged sessions requiring cryptographic CLI-hook integrity checks.