Safety, security & interpretability for advanced learners
71 resources · page 1 of 3
11x accused of claiming customers it doesn't have (TechCrunch)
by TechCrunch
TechCrunch report on AI sales startup 11x claiming customers it did not have.
9to5Mac: Figma explains why its AI kept copying Apple's Weather app
by 9to5Mac
9to5Mac report on Figma explaining why its Make Designs AI feature produced copies of Apple's Weather app.
Academ-AI
by Alex Glynn (University of Louisville)
Academ-AI paper documenting academic publications that contain undisclosed AI-generated text.
AI Alignment Forum
by AI Alignment Forum (Lightcone Infrastructure)
AI Alignment Forum: community blog for technical AI alignment research.
AI Hallucination Cases database (Damien Charlotin)
Damien Charlotin's database of court decisions worldwide involving AI-hallucinated content in legal filings, searchable by country and tool.
AI Incident Database: Report 3918
by Responsible AI Collaborative
AI Incident Database report 3918, an article on AI hallucinations in legal practice, as an example incident record.
AIID taxonomies
by Responsible AI Collaborative
AI Incident Database taxonomies page listing the classification schemes, such as CSET and GMF, applied to incidents.
AIxCC Competition Archive
AIxCC Competition Archive: DARPA AI Cyber Challenge records and code of finalist cyber reasoning systems.
Amazon scraps biased AI recruiting tool (Euronews/Reuters)
by Euronews
Euronews/Reuters report on Amazon scrapping an AI recruiting tool that showed bias against women.
Arup US$25M deepfake video-call fraud (Construction Briefing)
Construction Briefing report on an Arup employee transferring US$25 million after a deepfake video call.
AVID
by AI Vulnerability Database (AVID)
AVID: open database of vulnerabilities and failure modes in general-purpose AI systems, structured like CVE entries.
DMV Statement on Cruise LLC Suspension (California DMV)
California DMV statement (October 2023) suspending Cruise's driverless deployment and testing permits.
Center for AI Safety
Center for AI Safety: nonprofit doing AI safety research and field-building, creator of MMLU and Humanity's Last Exam.
Chicago Sun-Times AI reading list of nonexistent books (404 Media)
by 404 Media
404 Media report on the Chicago Sun-Times printing an AI-generated summer reading list with nonexistent books.
China GB 44721-2026 mandatory ADS safety standard (CnEVPost)
by CnEVPost
CnEVPost article (2026) on China's GB 44721-2026 mandatory safety standard for L3/L4 automated driving.
CNET found errors in more than half its AI stories (Talking Biz News)
Talking Biz News report on CNET correcting 41 of 77 stories written with an AI tool.
Cursor/Anysphere support bot invents a policy (CX Network)
by CX Network
CX Network report on Cursor's support chatbot inventing a false policy and the resulting backlash.
Deloitte partial refund to Australian government over AI errors (CTV)
by CTVNews
CTV report on Deloitte partially refunding the Australian government for a report with AI-generated errors.
Epoch AI
by Epoch AI
Epoch AI: research institute tracking trends in AI compute, data, models and their implications.
Zillow to shutter home buying business and lay off 2,000 employees (GeekWire)
by GeekWire
GeekWire report on Zillow closing its algorithm-driven home-buying business and laying off 2,000 staff.
Georgia Tech: AIxCC lessons at USENIX Security 2026
by Georgia Tech College of Computing
Georgia Tech news on its researchers' AI Cyber Challenge (AIxCC) lessons presented at USENIX Security 2026.
Haize Labs
by Haize Labs
Haize Labs: company providing red-teaming and evaluation services to move AI systems into production.
IBBIS (International Biosecurity and Biosafety Initiative for Science)
by International Biosecurity and Biosafety Initiative for Science (IBBIS)
IBBIS: international initiative strengthening biosecurity norms and tools such as DNA synthesis screening.
agent-scan (Snyk, formerly mcp-scan)
by Snyk
mcp-scan GitHub repo, now Snyk agent-scan: security scanner for AI agents, MCP servers and agent skills.
L1B3RT4S (Pliny jailbreak prompts)
by pliny
L1B3RT4S GitHub repo (Pliny): collection of jailbreak prompts for many AI models.
Law Gazette: judge alerts regulators over fake case authorities
by Law Gazette
Law Gazette report on a High Court judge referring lawyers to regulators over fake AI-generated case citations.
Global courts face rising tide of AI-generated fake cases and quotes (Malay Mail)
by Malay Mail
Malay Mail report on courts worldwide sanctioning AI-generated fake cases and quotes in legal filings.
MAST
by multi-agent-systems-failure-taxonomy
MAST GitHub repo: taxonomy and dataset of failure modes in multi-agent LLM systems.
Meta PurpleLlama / CyberSecEval
by Meta Llama
Meta PurpleLlama GitHub repo with CyberSecEval and Llama Guard, tools to assess and improve LLM security.
Transformer Circuits Thread
by Anthropic
Transformer Circuits Thread: Anthropic interpretability research series reverse-engineering transformer models.
Neuronpedia
by Neuronpedia
Neuronpedia: open-source interpretability platform for exploring sparse autoencoder features and circuits.
NTSB investigation HWY18MH010 (Uber ATG crash)
by US National Transportation Safety Board (NTSB)
NTSB investigation HWY18MH010: the 2018 Uber ATG automated test vehicle fatal crash in Tempe.