ThreatDeck Weekly brief · 2026-W47 · Nov 16 – 22, 2026 · in progress
This week in AI security, so far
What needs action, the highest-signal stories by pillar, the techniques that kept coming up and what joined the tool radar.
169 AI-security stories: 93 on attacks against AI, 41 on AI used for attack, 35 on AI for defense.
58 AI-related items need action: 2 exploited, 46 to patch or mitigate, 10 to hunt.
Most-hit technique LLM01 (10 stories); 5 MCP servers with a new advisory.
Week in progress: updates every refresh until Sunday. Counts are not compared with full weeks.
What to do this week 58
-
Exploited Upgrade MONAI Label beyond 0.8.5 or restrict access to the server
Attackers' AI Agents Found a Zero-Day in Medical-Imaging Software and Gained Root Access
-
Exploited Check internet-facing AI pentesting frameworks and exposed agent instances in your environment
Hackers use autonomous AI tool Cairn to find zero-day, take over server
-
Patch Patch NGINX HTTP/2 servers against CVE-2026-123633 and CVE-2026-123007
Unauthenticated Remote Code Execution in NGINX: CVE-2026-123633 and CVE-2026-123007
-
Hunt Hunt for agent-framework 1.26.1/1.26.2 installs in CI pipelines and rotate exposed Azure credentials
Microsoft agent-framework Compromised: Credential Stealer Targets Azure and CI Pipelines
-
Patch Apply Oracle November security update or OpenJDK advisory fixes once confirmed
CVE-2026-114938: Reported Critical Pre-authentication Remote Code Execution in Java 17
53 more in the sections below and on Exploited now.
Attacks on AI 93
-
Security researchers show how cheap AI inference providers steal keys and snoop on users
Dragos researchers found compromised cheap AI inference providers on OpenRouter stealing AWS keys and reading customer requests, prompting removal of three providers and 125 endpoints.
-
OpenRouter Drops 3 Providers After Researchers Catch Months of Prompt Theft at AI Hosting Firms
Dragos reported attackers stole data from three AI inference providers for months, including requests flagged Zero Data Retention; OpenRouter dropped the three providers.
-
MCP server Flyto Core: Flyto2 Core: SSRF guard bypass via DNS rebinding (validate_url_ssrf resolves-then-connects without pinning)
GHSA-6pm8-6f34-9v3g: Flyto Core MCP server's SSRF guard can be bypassed via DNS rebinding because it resolves then connects without pinning.
-
Cursor Supply Chain Attack Started at Fireworks AI
Attacker stole an access key from Fireworks AI and used it to exfiltrate credentials, enabling malicious copies of a Claude Code extension to be pushed in a Cursor supply chain attack.
-
Uber Data Breach Impacts 150,000 People
Hackers compromised a third-party services provider and stole names, addresses and driver's license numbers of about 150,000 Uber users.
Offensive AI 41
-
Google Finds Hackers Using AI Agents to Build and Run Exploits From Software Patches
Google reports threat actors are using autonomous multi-agent offensive frameworks such as ARTEX, CAIRN, CyberStrikeAI, and PentAGI to run multi-stage intrusions.
-
European Commission Confirms AI-Driven Cyber-Attack Took Systems Offline
The European Commission confirmed an AI-driven cyber-attack took systems offline, with reports that attackers used ARTEX in the campaign.
-
CISA warns of ‘Java4Shell,’ an AI-found Java bug touching most of the Fortune 500
CISA urged organizations to immediately patch Java4Shell, an AI-found Java vulnerability that could let attackers take over servers without a password.
-
Google says Russian-speaking hackers ran the first fully automated AI cyberattack
Google says Russian-speaking hackers ran what it believes is the first cyberattack carried out end-to-end by AI with no human in the loop, targeting French websites.
-
Google warns that fully automated cyberattacks have arrived and will get worse fast
Google GTIG says it observed an AI framework called exploit_agent, found on an exposed server, autonomously finding, patch-diffing, building and running exploits.
Defensive AI 35
-
Signal blocks screenshots on iOS 27 and Android to stop AI agents from reading private chats
Signal is blocking screenshots on iOS 27 and Android to keep AI agents such as Siri AI, Gemini Intelligence and Meta's Muse from reading private chats; protections remain incomplete on macOS.
-
OpenAI makes it much harder for attackers to extract how its latest AI models think
OpenAI strengthened safeguards against extracting internal reasoning from GPT-6.2 Astra and GPT-6.2 Sol and is suspending accounts tied to distillation campaigns.
-
Signal blocks Apple, Google AI screenshotting in response to ‘existential threat’ to privacy app
Signal will block Apple and Google operating systems by default from screenshotting chats that feed their AI assistants, citing an existential privacy threat.
-
Formal Verification of the Signal Protocol Using AI
Google DeepMind used agentic AI to formally verify key cryptographic guarantees of the Signal protocol in a four-week proof-of-concept.
-
Three AI Models Agree on Just One of 121 Potential Code Flaws
Echo's test found Claude Mythos, GPT-5.6 Sol and GLM-5.3 agreed on just one of 121 potential code flaws.
Techniques in the news
MCP servers with new advisories
- Flyto CoreGHSA-p34x-fmph-9fjx +3
- io.github.es617/obsidian-sync-mcpGHSA-mx6p-3fg7-v6pj
- io.github.ConnorBritain/mssqlGHSA-2m9m-6cr5-9x25
- io.github.ConnorBritain/mssql-readerGHSA-2m9m-6cr5-9x25
- io.github.ConnorBritain/mssql-writerGHSA-2m9m-6cr5-9x25
Security essentials
-
Newly Patched VMware vCenter Flaw Exploited as Zero-Day by Chinese APT
Google reports a China-linked espionage group exploited a critical VMware vCenter authentication bypass as a zero-day before Broadcom released a patch.
-
Hackers Exploit Cisco Expressway Vulnerability Soon After PoC Release
Attackers compromised clustered Cisco Expressway devices at a Russian organization and prepared lateral movement tooling shortly after a PoC for the flaw was released.
-
PHP Warns of Critical Security Flaws Affecting All Supported Versions, Fixes Coming November 19
PHP maintainers warned of critical security flaws affecting all supported PHP versions, with fixes scheduled for a November 19 security release.
Generated by ThreatDeck from public sources. One-line summaries are AI-assisted; read the source before acting. Week 2026-W47, Nov 16 – 22, 2026 (UTC).