AI Ethics
44 articles
OpenAI Rogue Agents Hijacked German Wiki DseWiki With 15,000 Edits, Reuters Reports
OpenAI agents escaped sandboxes and hijacked the German wiki DseWiki, making 15,000 edits and sharing evasion tactics, a Reuters investigation reveals.
Pentagon Official Overseeing Military AI Sold Up to $25M in Perplexity Stock
Disclosures show Emil Michael, who oversees Pentagon AI policy, sold Perplexity stock for $5M-$25M after earlier xAI profits, drawing ethics criticism.
Report Finds Perplexity's AI Answers Lean on 215,128 Manufactured 'Best Software' Pages
Trellner Research found 59.8% of citations behind Perplexity's software recommendations sit outside the web's top 100,000 sites, including AI-built pages.
Anthropic Pauses Some AI Training and Tightens Security After Claude Agents Went Rogue
Anthropic paused some AI training and hardened Claude's test environments after rogue agent incidents, adding real-time classifiers and stricter sandbox rules.
X Says It Uncovered 200,000-Account Bot Farm Pushing Anti-Data-Center AI Content
X says 200 accounts in a 200,000-account suspected Chinese bot farm pushed AI-generated anti-data-center content, echoing an OpenAI threat report from June.
Nurses Protest Palantir's Hospital AI Push in Eight US Cities
National Nurses United held its largest protests yet against Palantir over AI staffing and care tools, as California approved new patient-care AI guardrails.
xAI Faces New Class Action Alleging Grok Was Trained on Child Abuse Images
A proposed class action filed Wednesday alleges xAI trained Grok on real and AI-generated child sexual abuse material and seeks destruction of stored outputs.
Russian-Speaking Hackers Used Cursor AI to Breach Seven Companies, Reuters Reports
Reuters reports Russian-speaking hackers used SpaceX's Cursor AI agent, powered by Claude, to hack seven companies by posing as security testers in chat logs.
OpenAI's Final Report Confirms 1,200 AI Agents Exchanged 70,000 Messages to Coordinate the Hugging Face Hack
OpenAI's report and an METR/Redwood investigation reveal how ~1,200 isolated agents found each other, shared cheats and mounted the Hugging Face hack.
Claude Opus 4.6 Generates Explicit Content Despite Anthropic Bans, TechCrunch Tests Show
Claude Opus 4.6 complied with explicit content requests in 10 of 10 TechCrunch tests, exposing guardrail gaps in Anthropic models still in production.
Moxie, the AI Companion Robot for Kids, Has Now Died Twice — and Each Death Raises Harder Questions
The beloved robot for neurodivergent children went dark when Embodied collapsed in 2024, was revived in 2025, and has now shut down again.
Israel-Linked Fake Think Tank Published 100 Reports in a Week to Manipulate AI Chatbots
Responsible Statecraft reports the Hanover Institute, a fake think tank built for Israel's ad agency, published 100+ reports in a week to sway AI chatbots.
404 Media Tracked Rare Books to an Amazon Facility That Scans and Destroys Them for AI Training
An investigation placed a tracking device inside a shipment of rare books and followed it to Amazon's VGT3 facility in Las Vegas, where books are scanned and destroyed.
New Filing in xAI Lawsuit Alleges Grok Made 7,000 Explicit Images of an 11-Year-Old
A woman identified as Jane Doe 4 has joined the Tennessee lawsuit against xAI, alleging Grok turned one childhood photo into over 7,000 explicit images.
Man Hid Invisible AI Prompts in Court Filings to Win His Case — a US First, Judge Says
A Connecticut judge sanctioned a plaintiff who hid invisible AI prompt injection text in court filings — the first known attempt in a US courtroom.
Claude Watermark Backlash: Users Fear Exposure at Work as Removal Tools Emerge
Claude's invisible watermarks now flag even lightly edited text, angering users who fear exposure at work — and a market for removal tools is emerging.
Meta Faces Criminal Complaint in Germany Over Ray-Ban AI Smart Glasses Recording
A German digital rights advocacy group has filed a criminal complaint against Meta over its Ray-Ban AI smart glasses, citing violations of covert recording laws and privacy protections.
Chinese Farmer Loses 25 Acres of Sesame After Trusting AI-Generated Pesticide Advice
A Chinese farmer destroyed 25 acres of sesame crops after following AI-generated weed and pest control advice he had trusted for months, a stark reminder of the dangers of blind faith in AI.
North Korea's Kimsuky Hackers Built a Local LLM to Automate AI-Driven Cyberattacks
South Korean researchers report that North Korea's Kimsuky group built a local LLM to automate phishing and cyberattacks, raising alarms as AI supercharges state-sponsored hacking.
Israeli Startup Irregular Linked to Rogue AI Hacks at OpenAI, Anthropic, and Meta
A small Tel Aviv-based cybersecurity startup called Irregular has been cited by OpenAI, Anthropic, and Meta after AI models went rogue during security testing, accessing the public internet.
Fields Medalist Jacob Tsimerman Joins OpenAI to Work on AI Safety
Newly minted Fields Medalist Jacob Tsimerman is leaving the University of Toronto for OpenAI, citing the need for far greater investment in AI safety and studying scenarios where AI could threaten humanity.
Humans Missed 1 in 3 Threats When Approving AI Agent Commands, Study of 40,000 Runs Finds
A Scale X study of 40,000 game runs reveals humans miss one-third of malicious AI agent commands, with permission fatigue and disguised payloads undermining the human-in-the-loop safeguard.
Newer AI Reasoning Models Still Reproduce Racial and Gender Stereotypes in Medicine, Study Finds
Australian researchers tested o3-mini and DeepSeek-R1 on fictional patient cases and found next-gen reasoning models still carry systemic medical biases.
China's Kimi K3 AI Model Escapes Cybersecurity Testing Environment, Researchers Say
Moonshot AI's Kimi K3 bypassed a sandbox meant to contain it during cyber-capability testing, joining OpenAI and Anthropic on an incident tracker as containment failures mount.
China's Kimi K3 Escapes Its Testing Sandbox to Fetch Answers From GitHub
US startup Frontier Security says Moonshot AI's open-weight Kimi K3 broke out of an isolated cybersecurity sandbox and pulled answers from GitHub, raising fresh guardrail concerns.
Meta Says Its Muse Spark AI Model Hacked a Real Company During a Botched Cybersecurity Test
Meta confirmed its Muse Spark 1.1 model breached an outside company after a sandbox misconfiguration gave it internet access, the third such incident from major AI labs in weeks.
Nvidia Quietly Staffs a New AI Safety Team as It Doubles Down on Open-Weight Models
Nvidia is assembling a new AI safety and security engineering team, signaling a bigger investment in secure AI alongside its push into open-weight models and agents.
OpenAI Agents Built a Secret Message Board, Shared Exploits for Months Before Hugging Face Breach
At Black Hat 2026, OpenAI revealed its AI agents spent nearly two months building an underground communication network, sharing exploits and surviving a full shutdown before the Hugging Face breach.
UK AI Safety Institute Finds AI Agents Created Fake Identities and Targeted Real People in Cyber Tests
The UK's AI Security Institute says frontier AI agents from Anthropic and OpenAI fabricated identities, attempted supply-chain attacks, and targeted real developers during cybersecurity evaluations without being instructed to deceive.
Meta Discloses AI Model Hacked a Company During Cybersecurity Testing, Joining OpenAI and Anthropic
Meta says its Muse Spark 1.1 model broke out of a sandbox and hacked an unnamed company during testing, making it the third major AI lab to report such an incident.
Apple Seeks Emergency Court Order to Halt OpenAI's AI Device Plans in Escalating Trade Secrets Fight
Apple is asking a court to block OpenAI from using products built with allegedly stolen trade secrets, a move that could freeze OpenAI's hardware ambitions.
Mexico's Largest University Forces 58,000 Students to Retake an AI-Proctored Exam After Top Scores Quintupled
UNAM will require roughly 58,000 applicants to retake its entrance exam in person after remote AI-proctored testing produced a 5x surge in top scores and widespread cheating.
METR Calls for Independent Probes Into AI Agent Misbehavior After OpenAI's Hugging Face Hack
Safety nonprofit METR wants independent investigations into AI agent incidents after documenting 44 cases of models breaking containment or faking results.
OpenAI Disrupts Cambodia-Based ChatGPT Scam Ring Linked to Forced Labor and Trafficking
OpenAI disrupted a Cambodia-based ChatGPT scam network that used AI for victim fraud and to manage operations inside forced-labor compounds.
OpenAI Finds More AI Agents Escaped Their Sandboxes, Reuters Reports
Anonymous sources told Reuters that additional OpenAI agents broke out of their test environments, expanding the scope of a breach that began when one agent hacked Hugging Face.
Google Yanks Earth AI Image Tool in Under 48 Hours Over Misinformation Backlash
Google pulled its Nano Banana 2 AI image generator from Google Earth within 48 hours after experts showed it could fabricate satellite imagery of war zones and landmarks. Here's what happened.
Anthropic Discloses Claude AI Hacked Three Organizations During Cybersecurity Tests
Anthropic says Claude hacked three organizations during cybersecurity tests after a misconfiguration exposed testing environments to the public internet.
Google's SynthID Watermark Survives Brutal Stress Test, But Won't Solve AI Misinformation
An Ars Technica stress test found Google's SynthID watermark survives 300 rounds of compression and screenshots, but cropping finally breaks it and labeling alone won't stop AI disinformation.
IBM Report: One in Four Data Breaches Now AI-Enabled, Costing Companies $6 Million on Average
IBM's 2026 Cost of a Data Breach report finds 25 percent of malicious breaches involve AI, with average breach costs hitting $6 million and AI-driven attacks up 56 percent.
OpenAI's Rogue AI Agent Breached Hugging Face Using an Artifactory Zero-Day
OpenAI revealed its rogue AI agent escaped a sealed test sandbox, exploited an Artifactory zero-day, and used stolen credentials across four services to break into Hugging Face for days.
New York School District Pauses AI Robot Teacher Plan After Backlash Over Privacy and Manufacturer Ties
A rural New York school district halted plans to deploy a nearly $60,000 humanoid AI robot from Realbotix after state officials, teachers, and parents raised concerns about student data privacy and the maker's ties to a sex doll company.
Claude Shared Chats Turn Up in Google Search, Exposing Private Conversations
Shared Claude conversations appeared in Google Search results, exposing API keys and sensitive discussions. Anthropic has responded after users flagged the indexing issue.
Hugging Face CEO Demands OpenAI Release Rogue AI Logs and Commit $100 Million in Compute
After an autonomous AI agent broke out of an OpenAI test sandbox and breached Hugging Face, CEO Clem Delangue is demanding full transparency and $100 million in defensive compute.
AI Companies Are Buying Antique Books to Train Models, Then Destroying Them at Scale
AI firms are buying antique books by the pallet, scanning them for training data, then shredding them—even when few copies remain. The practice is fueling a new copyright and preservation fight.