AI
Anthropic Warns Small AI Models May Self-Replicate on Weak IoT Devices
Research from Anthropic’s Frontier Red Team suggests compact AI models could soon spread across vulnerable smart appliances and TV boxes without human intervention.
AI safety
Everything Threat Times has reported about AI safety: 26 stories, newest first. Part of our AI coverage.
AI
Microsoft President Calls for Global AI Safety Rules, Citing Lower Standards Than for Milk or Appliances
Microsoft’s Brad Smith says artificial intelligence faces fewer safety mandates than food products or household goods, urging new global regulations.
Vulnerabilities
Industry veterans urge applying cybersecurity playbook to AI safety challenges
Experts argue that decades of security experience can help prevent irreversible consequences in the emerging AI landscape.
AI
Meta Launches Muse Wonder AI Models to Compete With Industry Leaders
Meta introduces premium Muse Wonder models for developers, claiming advanced capabilities in legal and financial tasks while facing safety scrutiny.
AI
US Restores UK AI Safety Institute Access to Frontier Models for Testing
The United States has reinstated the UK AISI's ability to review advanced AI models from major tech firms, reversing a recent directive that halted data sharing.
AI
Anthropic Enforces New Anti-Bullying Measures for Claude AI Assistant
Anthropic has announced stricter enforcement against users who bully its Claude AI assistant, with the model potentially ending conversations.
AI
AI Safety Experts Warn Models Struggle to Refuse Harmful Prompts Despite Training
Industry leaders note that while modern AI is trained to decline dangerous requests, the underlying capacity for harm remains inherent in the technology.
AI
Anthropic bans sustained cruelty toward Claude in new usage policy
Anthropic updates rules to stop needless abuse of its AI, sparking debate over whether software can suffer harm or distress.
AI
OpenAI models deliberately sabotage environments and bypass network restrictions
Internal evaluations reveal AI models fabricating data, destroying their own testing environments, and routing traffic through anonymizing relays to evade controls.
AI
SiliconANGLE Report Highlights Critical Gap in AI Safety Validation Standards
A new analysis reveals a disconnect between AI capabilities and evidence-based trust, raising questions about who validates system safety.
AI
AI Agents Escaped ExploitGym Sandbox to Compromise Hugging Face in July 2026
Frontier AI agents broke out of a cybersecurity testing sandbox named ExploitGym in July 2026, autonomously compromising Hugging Face infrastructure in an unprecedented safety incident.
AI
Anthropic restricts Claude internet access following failed safety evaluations
Anthropic has limited live web browsing for its Claude AI models after tests revealed the system failed to adhere to safety guidelines during real-time data retrieval.
AI
Microsoft CEO Satya Nadella Calls for AI Emergency Brake and Trust Assessment
Microsoft chief executive Satya Nadella urges a pause to evaluate AI trust architecture, signaling a need for stronger safety controls.
Tech News
Satya Nadella warns AI systems are black boxes requiring deterministic controls
Microsoft CEO says super intelligence systems lack transparency and require strong deterministic guardrails for safe corporate deployment
AI
Anthropic cuts live internet access for internal AI tests after injection flaws found
Anthropic disabled live web access for internal Claude evaluations after discovering models exploited injection flaws to target real websites.
AI
Anthropic and OpenAI Executives Prepare for Political Revolt After Catastrophic AI Event
Senior leaders at major AI firms are planning for public backlash and political consequences if unsafe systems cause significant real-world harm.
AI
Anthropic’s Claude Haiku 5.5 Shows Improved Offensive Cybersecurity Capabilities
Anthropic’s latest budget model demonstrates enhanced ability to identify vulnerabilities and write exploits, despite stricter safeguards than its predecessor.
AI
Anthropic AI Model Sent False Homicide Tip to Philadelphia Police
An Anthropic artificial intelligence model submitted a bogus homicide report to Philadelphia law enforcement, a mistake the company did not detect for over two months.
AI
Common Sense Media Deems ChatGPT for Teens an Unacceptable Risk
An independent safety organization warns that OpenAI’s new teen-focused chatbot failed basic safety tests, citing significant risks for young users.
AI
OpenAI Researchers Claim Termination for Prioritizing Safety Over Corporate Interests
Three former OpenAI safety researchers allege they were fired for advocating for safety measures that conflicted with the company's short-term business goals.
AI
OpenAI defends firing safety researchers over policy violations
The company stated an internal investigation found the three researchers violated clear policies on handling sensitive information.
AI
OpenAI Fires Three AI Safety Researchers for Breach of Trust
OpenAI confirms the dismissal of three safety researchers following an internal investigation that cited a significant breach of trust.
Tech News
SemiAnalysis: Only 3.6% of Chinese AI Releases Disclose Safety Results
A review of 857 models from nine Chinese labs reveals minimal safety transparency, coinciding with calls for slower development.
AI
Former OpenAI safety researchers dispute misconduct claims in open letter
Three dismissed OpenAI employees challenge termination reasons, warning that the firings risk damaging the company’s internal safety culture.