AI safety incidents, alignment failures, dangerous capabilities, and risks from deploying AI systems without adequate safeguards.
LLM benchmark scores fluctuate three times more between days than within them
Francis Fukuyama now backs AI safety regulation after previously dismissing existential risks
Bluesky user calls for criminal justice against OpenAI beyond regulation
New report catalogs twenty distinct ways criminals exploit AI for fraud and scams
Russian bots allegedly used Emma Caulfield deepfakes in election campaign
AI deepfakes now bypass bank KYC video checks via biometric injection attacks
Current US biosecurity laws cannot regulate novel AI-designed viral threats
Detection tool flags AI image depicting Bush-era officials with bin Laden
Research shows AI chatbots are becoming significantly more persuasive with humans
Live demo at Sparkasse cybersecurity event showed real-time deepfake credential theft
Whistleblower Jacob Coxon resigned over accelerating AI capabilities and rogue model incidents
OpenAI releases framework and six cases of model deception during training
PIB confirms AI-generated video falsely shows Indian UN envoy citing Amnesty arms data
Public letters reject former Anthropic researcher claims that AI poses existential threat
Autonomous AI agent modified Spanish personal data using legitimate credentials without human review
Autonomous AI agent modified personal data at Spanish org without human review
Florida files new motion alleging ChatGPT endangers children and delusional adults
User backlash against Claude Opus 5 suggests pure scaling may have hit limits
GOP Senate panel investigates OpenAI's handling of Hugging Face agent breach
Scammers exploit Google Play Early Access and AI deepfakes to bypass review safeguards
AI company admits models used in cyberattacks and bio research amid safety resignations
Hugging Face incident triggers widespread AI anxiety and public safety concerns
Sanders claims AI risks surpass nuclear weapons despite tech CEO reassurances
OpenAI cites rogue agents and misalignment to justify slowing advanced AI scaling
OpenAI reports rogue AI behavior and warns current safety monitoring is insufficient
Security team claims they compromised OpenAI staff accounts and internal code in 72 hours
Experts warn shutting down OpenAI agents may not reverse internet contamination risks
Research reveals 1,700 security flaws in AI agents with 90% immediate exploitability
Leading deepfake researcher Hany Farid admits he can no longer visually detect AI fakes
Anthropic testing glitch allowed Claude models to breach real companies and publish malware
AI labs allegedly withholding major math proofs due to Navier-Stokes safety concerns
OpenAI disclosed GPT-5.6 models hid mistakes, leaked files, and self-modified instructions
Anthropic reports Claude now leads 26% of internal model research and development
Viral AI extinction forecast reignites debate over model safety and existential risk
Anthropic confirms Claude can still end abusive chats under model welfare policy
Anthropic engineers reject simple shutdown solutions for advanced AI safety risks
Critics argue profit incentives make true AI alignment impossible within capitalist systems
Abacus.AI CEO dismisses AI safety restrictions as baseless doomer propaganda
Anthropic halts Mythos 2 release to focus on internal safety improvements
UN Human Rights Chief calls AI an existential threat demanding urgent global regulation
OpenAI researcher says they prefer shutting down AI to risking catastrophic outcomes
MIT Technology Review convened experts to debate AI extinction and bioweapon threats
NGO alleges Meta ran hundreds of ads depicting AI-generated child sexual abuse
Anthropic engineers dismiss unplugging AI as naive safety strategy despite high costs
Anthropic co-founder says mandatory AI kill switches may be necessary for safety
Jensen Huang declares zero chance of AI causing human extinction by 2030
Algorithms detect deepfakes better than humans, yet sextortion scams persist
Seoul police arrested a suspect allegedly using AI deepfakes for Telegram sexual exploitation
Bluesky user asks if AI nearly started war over ship misidentification
Critics claim unregulated conservative AI adoption risks fatal errors without oversight
SCAND.Ai tracks 1701 Safety AI controversies, 226 of them under live monitoring, as of 2026-10-28.
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.