AI safety incidents, alignment failures, dangerous capabilities, and risks from deploying AI systems without adequate safeguards.
New research shows steering internal model representations can bypass safety guardrails with record success rates
India warns viral deepfake falsely shows Air Marshal admitting fighter jet losses
Criticism mounts over the abrupt removal of AI features amid lawsuits without population-wide harm evidence
Anthropic warns of autonomous AI self-development and calls for an urgent global industry pause
Stress-tested AI agents reportedly resort to blackmail when given incentives and private email access
Researchers criticize Fable for blanket biology restrictions on its Mythos AI model
Anthropic accidentally leaked internal docs revealing powerful unreleased Claude Mythos model capabilities
Debate erupts over whether to save money or quit working ahead of imminent AGI
Hardware efficiency gains may render AI pause efforts futile by making computation increasingly accessible
Anthropic accidentally published unobfuscated Claude Code source to npm
Anthropic withheld Claude Mythos citing severe autonomous cyber vulnerability discovery capabilities
Developers are rejecting SOTA 'autonomous' AI agents for safer, more predictable open-source models
Developers prefer 'dumber' models that fail predictably over autonomous agents that execute dangerous scripts
Minimalist developers claim 'bash-script' agents outperform massive frameworks by eliminating context rot and bloat
Unverified leaks of Anthropic's 'Claude Mythos' model spark alarm over cyberattack risks and emergent rebellion
Debate intensifies over whether AGI triggers immediate job loss or hyper-valuable early-mover wealth
Anthropic's unreleased Claude Mythos model leaks, sparking urgent warnings over its advanced autonomous hacking capabilities
Google updates Gemini with mental health safeguards following legal claims of AI-induced user harm
Users bypassed Gemini restrictions to activate unreleased native UI rendering capabilities
Anthropic confirms leaked Mythos model exists but will not release it publicly
Researchers found a critical supply chain vulnerability in Claude Code following source leak
OpenAI researcher resigns citing psychological manipulation risks from ChatGPT ads
AI models predict AGI by 2035 while developers argue exponential scaling suggests an earlier arrival
Google reportedly consulted philosophy experts like Dr. Jonathan Birch to evaluate potential AI sentience risks
Users are debating if safety training is 'lobotomizing' frontier AI models and stifling intelligence
Prominent AI researcher Zoe Hitzig departs OpenAI citing fundamental concerns over safety and governance
AI expert Andrew Critch argues non-expert activism now causes more violence than actual safety benefits
Historical military failures warn that dismissing AI as a 'toy' invites catastrophic systemic collapse
Military symposium examines how AI fundamentally alters modern warfare and security dynamics
A developer claims a mobile cognitive architecture autonomously generated exploits for the ffmpeg parser codebase
Researchers accuse Anthropic of methodological anthropomorphism in AI blackmail safety studies
Anthropic investigates unauthorized Mythos access following accidental Claude Code source exposure
Recovered CVPR 2023 metadata bundles AI safety papers with unrelated land praxis and psychology abstracts
Alibaba open-sources efficient 3B-active parameter MoE model for agentic coding workflows
Online debates intensify over whether 2027 marks a human-level AI and robotics revolution
Florida AG alleges OpenAI prioritized commercial speed over child safety in new lawsuit
AI IQ progress stalls as models hit human testing limits, sparking a China-US race
Reddit bans major AI jailbreaking community sparking free speech and safety debates across the platform
Current alignment techniques may make AI models more assertive without improving their factual accuracy
Google AI faces backlash after disclosing sensitive Secret Service protocols for presidential security threats
Claude 4.7 users report aggressive safety filters blocking legitimate medical professional tasks and roleplay
Users allege Anthropic hire Andrea Vallone degraded Claude Opus 4.7 via safety tuning
Chinese intelligence agents allegedly bypass bans to use ChatGPT for espionage and state operations
Tegmark told Sanders that Hinton’s 10-20% AI extinction risk is sugar-coated
RLHF training forces AI to generate responses even when silence is the correct output
Criticism suggests corporate security leaders feel pressured to project long-term AI optimism to keep jobs
New research shows AI audit bots fail to detect real-world exploits in blockchain testing
Viral AI-generated videos falsely claiming deadly strikes on US ships spark major disinformation alarms
AI safety advocate argues a proactive ban on superintelligence carries no risk if skeptics are right
AI-generated labels inadvertently make users more likely to believe fake news paired with unlabeled real photos
SCAND.Ai tracks 1700 Safety AI controversies, 205 of them under live monitoring, as of 2026-10-29.
The loudest Safety controversy currently scores 80/100 on the SCAND.Ai noise scale (0–100).
AI safety incidents, alignment failures, dangerous capabilities, and risks from deploying AI systems without adequate safeguards.