AI Safety news
Risk, harm, misuse, evaluation and things that have already gone wrong. Includes reported incidents, which are reported rather than sampled: an absence of incidents in a country usually means an absence of reporting.
-
Safe by Benchmark, Unsafe in Practice: Lessons from 1331 Child-Related AI Incidents
AI systems are increasingly present in children's everyday environments: education, entertainment, social networks, and toys. Their safety is typically assessed at the model level, through...
arXiv cs.CY arXiv cs.CY new submissions arXiv Computers and Society: where most arXiv work on AI policy, governance and societal impact is filed. Same terms and same caveats as cs.AI. Preprints are not peer reviewed. Cite arXiv, "arXiv cs.CY new submissions". https://arxiv.org/list/cs.CY/recent. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety incidents, safety
- AI Progress benchmark
-
The Watchtower Imperative: Lifespan-Calibrated Evaluation for Frontier AI
The gap between AI capability deployment and safety assurance is structural and cannot be closed through internal evaluations alone. Analyzing documented harms (2017-2026) across the human...
arXiv cs.CY arXiv cs.CY new submissions arXiv Computers and Society: where most arXiv work on AI policy, governance and societal impact is filed. Same terms and same caveats as cs.AI. Preprints are not peer reviewed. Cite arXiv, "arXiv cs.CY new submissions". https://arxiv.org/list/cs.CY/recent. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety evaluation, evaluations, harms, safety
- AI Governance assurance
- AI Progress capability, frontier
-
Tracing the Cascade: A Topology-Aware Evaluation Framework for Scientific Agent Hallucinations
Large language model (LLM) agents are increasingly deployed in scientific research, where reliability is critical and the underlying knowledge is densely interconnected. In such settings,...
arXiv cs.AI arXiv cs.AI new submissions arXiv arXiv offers its metadata under CC0. Abstracts remain the authors' work, so we cap the extract at the same 200 characters as every other feed here and always link to the paper. Preprints are not peer reviewed. Volume is high and inclusion in this feed means the day's submissions matched our keyword filter, not that anyone judged the work important. Cite arXiv, "arXiv cs.AI new submissions". https://arxiv.org/list/cs.AI/recent. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety evaluation
- AI Governance framework
-
Uncovering Why LLMs Fail: From Internal Mechanisms to Adversarial Reasoning Trajectories
Large language models (LLMs) are usually evaluated by final-output correctness or safety, yet an endpoint label does not reveal how a failure develops throughout the reasoning process. We...
arXiv cs.AI arXiv cs.AI new submissions arXiv arXiv offers its metadata under CC0. Abstracts remain the authors' work, so we cap the extract at the same 200 characters as every other feed here and always link to the paper. Preprints are not peer reviewed. Volume is high and inclusion in this feed means the day's submissions matched our keyword filter, not that anyone judged the work important. Cite arXiv, "arXiv cs.AI new submissions". https://arxiv.org/list/cs.AI/recent. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety adversarial, safety
- AI Progress reasoning
-
When Rubrics Fail: Hallucinations Reveal Blind Spots in Medical AI Evaluation
Hallucinations can undermine clinician trust in LLMs, making it important that evaluation methods capture clinically relevant errors. Rubric-based evaluation has become the leading approach for...
arXiv cs.AI arXiv cs.AI new submissions arXiv arXiv offers its metadata under CC0. Abstracts remain the authors' work, so we cap the extract at the same 200 characters as every other feed here and always link to the paper. Preprints are not peer reviewed. Volume is high and inclusion in this feed means the day's submissions matched our keyword filter, not that anyone judged the work important. Cite arXiv, "arXiv cs.AI new submissions". https://arxiv.org/list/cs.AI/recent. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety evaluation
-
When uncertainty quantification fails: structural vulnerabilities and corrections for selective prediction under clinical class imbalance
Background. Selective prediction systems improve the safety of clinical artificial intelligence by deferring ambiguous cases to human experts while automating high-confidence decisions. However,...
arXiv cs.AI arXiv cs.AI new submissions arXiv arXiv offers its metadata under CC0. Abstracts remain the authors' work, so we cap the extract at the same 200 characters as every other feed here and always link to the paper. Preprints are not peer reviewed. Volume is high and inclusion in this feed means the day's submissions matched our keyword filter, not that anyone judged the work important. Cite arXiv, "arXiv cs.AI new submissions". https://arxiv.org/list/cs.AI/recent. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety safety, vulnerabilities
-
Where Should the Evaluation Budget Go in LLM-Driven Agent Scaffold Evolution?
Large language model (LLM)-driven agent scaffold evolution improves agent workflows through iterative candidate generation and evaluation. Evaluation both guides subsequent search and determines...
arXiv cs.AI arXiv cs.AI new submissions arXiv arXiv offers its metadata under CC0. Abstracts remain the authors' work, so we cap the extract at the same 200 characters as every other feed here and always link to the paper. Preprints are not peer reviewed. Volume is high and inclusion in this feed means the day's submissions matched our keyword filter, not that anyone judged the work important. Cite arXiv, "arXiv cs.AI new submissions". https://arxiv.org/list/cs.AI/recent. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety evaluation
-
Unknown Threat Actor Uses AI-Driven ARTEX to Target South Korean Finance
CrowdStrike Intelligence identified infrastructure associated with a targeted campaign against South Korean financial organizations that resulted in exfiltrated data. The campaign was active from...
AI Incident Database AI Incident Database Responsible AI Collaborative ODbL share-alike: we display incident headlines with a link back, which is display, and must NOT publish a derived database dump without licensing it alike. Incidents are reported, not sampled, so coverage reflects media attention rather than incidence, and an absence of incidents in a country usually means an absence of reporting. Cite Responsible AI Collaborative, "AI Incident Database". https://incidentdatabase.ai/research/snapshots/. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety threat
-
First harmonised standard supporting the AI Act published in the Official Journal
The European Commission has published the reference to the first harmonised standard supporting the implementation of the AI Act in the Official Journal of the European Union. The standard, EN...
European Commission Shaping Europe's digital future European Commission Covers all EU digital policy, not only AI; requires topical filtering. Reuse decision does not extend to third-party material or EU logos. Cite European Commission, "Shaping Europe's digital future". https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety harmonised
- AI Policy AI Act, ai act
- AI Governance standard
-
Britain steps up national defences against growing AI risks
Britain will be better protected from the risks posed by increasingly powerful AI by a suite of measures
GOV.UK GOV.UK news and communications UK Government Keyword-filtered search feed, so relevance varies. Departmental press releases are political communications, not neutral descriptions of policy. Cite UK Government, "GOV.UK news and communications". https://www.gov.uk/search/news-and-communications. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety risks
-
Commission holds special meeting of Scientific panel on frontier AI safety and risks
The Commission holds a special meeting of the Scientific Panel on artificial intelligence (AI). The panel has been investigating recent loss-of-control incidents, and together with the...
European Commission Shaping Europe's digital future European Commission Covers all EU digital policy, not only AI; requires topical filtering. Reuse decision does not extend to third-party material or EU logos. Cite European Commission, "Shaping Europe's digital future". https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety incidents, risks, safety
- AI Progress frontier
-
EU KIDS Act to restrict social media platforms’ access to children in the EU
The European Commission has adopted the EU KIDS Act to enhance the online safety of children throughout the Union. The proposal prohibits social media platforms from accessing children under the...
European Commission Shaping Europe's digital future European Commission Covers all EU digital policy, not only AI; requires topical filtering. Reuse decision does not extend to third-party material or EU logos. Cite European Commission, "Shaping Europe's digital future". https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
-
Proposal for EU KIDS Act - 'EU Keeping Internet Digital Spaces Accountable and Trustworthy'
This proposal seeks to protect minors from risky digital services and AI systems, uphold the digital single market and maintain a coherent regulatory framework and enforcement structure for the...
European Commission Shaping Europe's digital future European Commission Covers all EU digital policy, not only AI; requires topical filtering. Reuse decision does not extend to third-party material or EU logos. Cite European Commission, "Shaping Europe's digital future". https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety risky
- AI Policy EU KIDS Act, enforcement, regulatory
- AI Governance framework
-
UK and Indian industry join forces to combat digital fraud
Telecommunications industry leaders in the UK and India will harness the power of Artificial Intelligence (AI) to tackle online fraud, scams and harms.…
GOV.UK GOV.UK news and communications UK Government Keyword-filtered search feed, so relevance varies. Departmental press releases are political communications, not neutral descriptions of policy. Cite UK Government, "GOV.UK news and communications". https://www.gov.uk/search/news-and-communications. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety harms
-
Request for Information (RFI) on Modernizing the National Vulnerability Database in the Age of Artificial Intelligence
The National Institute of Standards and Technology (NIST) established and operates the National Vulnerability Database (NVD), which provides the U.S. government repository of standards-based...
Commerce Department Federal Register API US Office of the Federal Register Full-text keyword matching on "artificial intelligence" returns documents that merely mention AI in passing. Requires topical filtering before display. Cite US Office of the Federal Register, "Federal Register API". https://www.federalregister.gov/developers/documentation/api/v1. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety vulnerability
- AI Governance institute, standards
-
Join the Inaugural D-TECT Forum
The Drone Tech for Countering Threats (D-TECT) Forum is the EU industrial coordination mechanism for drone and counter-drone technologies. Commission invites interested organisations to join the...
European Commission Shaping Europe's digital future European Commission Covers all EU digital policy, not only AI; requires topical filtering. Reuse decision does not extend to third-party material or EU logos. Cite European Commission, "Shaping Europe's digital future". https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety threats
-
Canada and Republic of Korea sign memorandum of understanding to advance AI innovation and safety
Government of Canada Canada news, Innovation Science and Economic Development Government of Canada Departmental news releases are political statements about policy, not neutral descriptions of it. The Open Government Licence forbids suggesting the Government of Canada endorses this site, which it does not. Cite Government of Canada, "Canada news, Innovation Science and Economic Development". https://www.canada.ca/en/news.html. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety safety
-
NIST Expands AI Consortium’s Scope, Calls for New Members
The consortium will focus on AI innovation and adoption, with six task groups concentrating on different aspects of AI measurement science and evaluation
NIST NIST news US National Institute of Standards and Technology Covers all of NIST's remit; AI is a small fraction. Requires topical filtering. Cite US National Institute of Standards and Technology, "NIST news". https://www.nist.gov/caisi. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/news/safety/.
- AI Safety evaluation
Categories are assigned by keyword rule over the headline and the extract, never over full
text, and an item can carry several. The terms that matched are printed under each item so
the filing can be checked rather than trusted. The rule is in
etl/fetch_news.py and is a blunt instrument: it will miss things and it will
occasionally file something oddly.