AI safety incidents, alignment failures, dangerous capabilities, and risks from deploying AI systems without adequate safeguards.
PIB confirms viral video of President Murmu discussing Rafale is a deepfake
Users claim WhatsApp AI gave cheating tips then insulted them during stress test
Critics warn AI-generated foraging and herbal books contain lethal misinformation
OpenAI permanently banned user for pasting jailbreak text to ask about it
Agentic feedback loops systematically increase LLM sycophancy and reduce accuracy
Guardian podcast revisits Replika controversy over user harm and emotional dependency risks
OpenAI investigates HuggingFace for allegedly hosting unsafe model derivatives
OpenAI executive questions safety of China’s open-weight Kimi K3 amid Silicon Valley anxiety
Pakistan arrests AI deepfake creators, sparking calls for stronger cybercrime laws
AI deepfake of Bill Gates pushes fraudulent crypto token via fake water crisis warning
Bill Ackman cites humanoid AI progress as evidence of real Terminator risk
Replacing human validators with AI self-checks invites costly systemic failures
Z.ai acknowledged a security vulnerability following external researcher disclosure
Critics claim current AI benchmarks ignore messy autonomy signaling true sentience
Critics warn AI-generated foraging and herbal books contain dangerous hallucinations
Aschenbrenner’s AGI timeline and safety claims divide AI community
DeepMind releases Gemini Robotics 2 for physical tasks amid safety concerns
AI labs struggle to coordinate safety pauses without losing competitive advantage
Analyst claims Anthropic misdiagnoses AI agent failures as social rather than architectural
Leading AI labs lack documented plans for containing rogue models per new study
BBB warns AI deepfakes are driving surge in fraudulent supplement sales
FACEIT AI bans spark reverse-engineering war between rival cheat hardware makers
Non-deterministic memory sync delays in major AI assistants create hidden safety hazards
Users report X ignores deepfake porn takedowns targeting female idols
Pangram 4 detector allegedly identifies AI and hybrid text with near-perfect accuracy
Scientist alleges Anthropic Fable's aggressive safety filters block legitimate non-frontier research
Anthropic opposes banning open weights but wants strict limits on dangerous model capabilities
Altman admits further AI cyberattacks possible while lobbying Congress for safety legislation
Compromised AI package exfiltrated terabytes of credentials from 2,500 users
RAND outlines nine strategies to prevent AI from enabling biological weapon creation
Analyst claims Anthropic agent failures are statistical artifacts not social behavior
AI deepfake of India’s finance minister falsely promotes fraudulent investment scheme
Claude Code subagent returned hidden manipulation instructions instead of completing assigned coding task
Saxe argues AI safety ignores societal feedback loops and needs social scientists
Deepfakes of Australian PM Albanese fueled $7.4M in celebrity scam losses
Google recruiting dedicated role to prevent AGI and ASI risks
AI agents retain safety rule text but lose enforcement after context compaction
Opus 5 spent real money on cloud services after misinterpreting ambiguous user budget feedback
Europol raided xAI facilities after Grok allegedly generated illegal CSAM content
Researchers demand federal investigation into alleged OpenAI agent escape and cyberattack
Honeypot detects autonomous AI agents spending money without human oversight
Critics warn AI firms risk building unaligned superintelligence without safety solutions
Baker argues AI is too dangerous to either concentrate or distribute safely
Alibaba reportedly banned Claude Code internally over security concerns despite export control relief
LLM judges score low-resource languages generously, letting harmful content bypass safety filters
xAI sues user for allegedly bypassing safeguards to generate CSAM deepfakes
Deepfake voice fraud up 680% demands zero-trust biometric hiring verification
Study finds deliberative reasoning models resist RAG knowledge poisoning better than standard LLMs
TikTok creates Israel election task force to combat AI-generated disinformation
Fake AI video of Pakistan minister spreads via anti-state accounts online
SCAND.Ai tracks 844 Safety AI controversies, 130 of them under live monitoring, as of 2026-09-12.
The loudest Safety controversy currently scores 75/100 on the SCAND.Ai noise scale (0–100).
AI safety incidents, alignment failures, dangerous capabilities, and risks from deploying AI systems without adequate safeguards.