Meta reports that a non-invasive system decodes typed sentences from brain signals, and other teams have reconstructed viewed images from fMRI. The demonstrations are real, and they are narrower than the headlines. The question is how to tell the difference before the technology arrives.
What happened most recently, newest event first. Source-reported facts, editorial synthesis and inference are labelled on every entry, with the event date and the time we added it kept separate.
The Climate Council's report "Pollution Pipeline: Australia's Dirtiest Data Centres," dated 19 November 2026 on the page, says the largest data-centre proposals in Australia could add 39.7 million tonnes of climate pollution a year by 2035 if not matched by new renewables and storage. It says this is "almost as much pollution as our three dirtiest coal power stations emit today: Loy Yang A, Eraring and Bayswater." The page says there are 225 data centres lined up to connect to the main grid, "more than double the number a year ago," and that the Western Downs data centre proposal alone could add 8.3 million tonnes a year. Why it matters to the watch: the report tracks power and grid connections for data centres, which the AI-link rule covers as energy or grid stress for AI; the page itself does not mention AI.
llama.cpp advisories dated within the last 30 days (by advisory date): 13 flaws. Each flaw is a separate NVD or GHSA advisory record; duplicate CVE IDs are counted once. Examples from the advisories: (1) CVE-2026-126974: llama.cpp GGUF File Parsing Integer Overflow Remote Code Execution Vulnerability. (2) CVE-2026-126975: llama.cpp GGUF File Parsing Heap-based Buffer Overflow Remote Code Execution Vulnerability. (3) CVE-2026-126976: llama.cpp MTMD Multimodal GGUF File Parsing Integer Overflow Remote Code Execution Vulnerability. Severity mix: not checked; CVSS scores are not recorded in the CMS records. No flaw in this roll-up is listed in the CISA KEV catalogue or reported as exploited in the wild. Why it matters to the watch: these flaws are in AI software, which is the AI link under the AI-link rule.
MLflow advisories dated within the last 30 days (by advisory date): 9 flaws. Each flaw is a separate NVD or GHSA advisory record; duplicate CVE IDs are counted once. Examples from the advisories: (1) CVE-2026-118623: MLflow PickleEvaluationArtifact Deserialization of Untrusted Data Remote Code Execution Vulnerability. (2) CVE-2026-118627: MLflow SHAP load_explainer Deserialization of Untrusted Data Remote Code Execution Vulnerability. (3) CVE-2026-115886: MLflow: SSRF on unauthenticated MLflow servers via third-party scorer kwargs in the scorer invoke, register, and GenAI evaluate endpoints. Severity mix: not checked; CVSS scores are not recorded in the CMS records. No flaw in this roll-up is listed in the CISA KEV catalogue or reported as exploited in the wild. Why it matters to the watch: these flaws are in AI software, which is the AI link under the AI-link rule.
Other AI software advisories dated within the last 30 days (by advisory date): 7 flaws. Each flaw is a separate NVD or GHSA advisory record; duplicate CVE IDs are counted once. Examples from the advisories: (1) CVE-2026-114612: LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. (2) CVE-2026-123818: LMDeploy is a toolkit for compressing, deploying, and serving large language models. (3) CVE-2026-59158: Nuxt Ollama integrates the official Ollama JavaScript library into Nuxt applications. Severity mix: not checked; CVSS scores are not recorded in the CMS records. No flaw in this roll-up is listed in the CISA KEV catalogue or reported as exploited in the wild. Why it matters to the watch: these flaws are in AI software, which is the AI link under the AI-link rule.
vLLM advisories dated within the last 30 days (by advisory date): 33 flaws. Each flaw is a separate NVD or GHSA advisory record; duplicate CVE IDs are counted once. Examples from the advisories: (1) CVE-2026-115915: vLLM: Untrusted EPS inputs can trigger denial of service or Ghostscript RCE. (2) CVE-2026-115918: vLLM: Responses API bypasses trust_request_chat_template for chat_template_kwargs. (3) CVE-2026-115912: vLLM: Public multimodal SHM handle forgery reaches worker-side `pickle.loads()` with insecure serialization disabled. Severity mix: not checked; CVSS scores are not recorded in the CMS records. No flaw in this roll-up is listed in the CISA KEV catalogue or reported as exploited in the wild. Why it matters to the watch: these flaws are in AI software, which is the AI link under the AI-link rule.
Anthropic spent nearly $10 billion on compute and infrastructure in the third quarter of 2026. Why it matters to the watch: included under the cyber-attack exception, which covers vulnerabilities under active exploitation; the KEV entry does not link this to AI.
On 18 November 2026 the Governor's office published a release on frontier AI safety. The release says California's Government Operations Agency "consulted with" the expert group behind the report. The release lists "Power to cap the development and use of recursive self-improvement (RSI) to reduce the risk of catastrophic harm" and "giving the state authority to order an immediate shutdown in an emergency." It also lists "Expand safety incident reporting." The release defines RSI as "a potentially dangerous scenario in which an AI model can design its own successor without human input." Why it matters to the watch: the release proposes power to cap the development and use of recursive self-improvement and to order an immediate shutdown in an emergency, which bears directly on AI governance and safety.
A paper published 18 November 2026 in Frontiers in Bioengineering and Biotechnology (Biosafety and Biosecurity section) audits 100,000 SARS-CoV-2 NCBI BioSample records with 23 checks across five metadata layers. The abstract says: "Every record triggered at least one APBT relevant metadata exposure indicator," and 51.5% were classed High or Critical. It also says: "The findings do not demonstrate deliberate metadata poisoning." The authors call the risk an advanced persistent biological threat (APBT), borrowing the cyber idea of an advanced persistent threat for genomic surveillance. Why it matters to the watch: Table 1 of the paper lists the APBT "Role of AI" as "Generation of plausible synthetic sequences, metadata patterns, adversarial labels," which ties this biosurveillance data risk to AI.
CISA added CVE-2024-10441, an improper encoding or escaping of output vulnerability in Synology DiskStation Manager (DSM), to its Known Exploited Vulnerabilities catalog on 18 November 2026, with a federal remediation due date of 21 November 2026. The catalog entry requires vendor mitigations under CISA's BOD 26-04. The catalog records no known ransomware use. Why it matters to the watch: included under the cyber-attack exception, which covers vulnerabilities under active exploitation; the KEV entry does not link this to AI.
CISA added CVE-2026-40217, an unprotected alternate channel vulnerability in BerriAI LiteLLM, to its Known Exploited Vulnerabilities catalog on 18 November 2026, with a federal remediation due date of 2 December 2026. The catalog entry says the vulnerability is being exploited and requires vendor mitigations under CISA's BOD 26-04. The catalog records no known ransomware use. Why it matters to the watch: included under the cyber-attack exception, which covers vulnerabilities under active exploitation; the KEV entry does not link this to AI.
CISA added CVE-2026-90436, an unsafe reflection vulnerability in Apache Axis, to its Known Exploited Vulnerabilities catalog on 18 November 2026, with a federal remediation due date of 21 November 2026. The catalog entry requires vendor mitigations under CISA's BOD 26-04. The catalog records no known ransomware use. Why it matters to the watch: included under the cyber-attack exception, which covers vulnerabilities under active exploitation; the KEV entry does not link this to AI.
54 with an event in the past 7 days, showing the 12 newest. 163 published signals in the current snapshot. 163 signals: 156 reviewed by editors, 7 published automatically after an AI check. 34 sources are monitored for new items; each is checked before it is published. This is a curated selection, not a complete record of global events. See all on Latest →
A global perspective.
Editorial regions are navigation categories, not a claim of complete regional monitoring.
Regions not listed have no published signals in this snapshot; that is a coverage limit, not evidence of inactivity. 81 published signals have no established country and are listed without a region.