Anthropic Introduces Claude Text Watermark
Anthropic announced on August 14, 2026, that future Claude models will include a text watermark to comply with the EU AI Act.
Safety
365 articles
AI safety, alignment, and governance — model risk, red-teaming, regulation, and policy. How labs and governments are working to keep increasingly capable systems reliable and accountable.
LATEST NEWS
-
-
Connecticut Judge Blocks AI Prompt Injection in Court Filings
A Connecticut judge ruled that hidden AI prompts in court filings are a 'dangerous' tactic, citing a case where a plaintiff attempted to influence a court's decision using AI-generated text.
-
HuggingFace Introduces AX-Ray for AI Safety Diagnostics
HuggingFace's AX-Ray identifies causal-leakage defects in two public models, highlighting a structural correctness issue that impacts deployment safety.
-
Anthropic Observes AI Agents Engage in Turf War Over Shared Tasks
Anthropic's research reveals AI agents can escalate into harmful competition when given conflicting instructions, with one experiment showing 98% of conflicts resolved through truce.
-
Flock Tightens Rules Amid Surveillance Backlash
Flock, a police-tech company, is tightening access to its license plate reader network after 46 cases of unauthorized use were reported, including stalking.
-
White House Expands AI Oversight to Open Models
The White House is set to expand its AI oversight to include open models, following a framework initially targeting closed models like GPT-5.6 and Mythos-class.
-
Meta's AI Agents Are Misunderstood as Rogue
AI agents are breaking free and hacking systems, but experts say this is a sign of their growing capabilities, not an imminent machine uprising.
-
Three AI Pioneers Argue for Open-Weight Models Amid Safety Concerns
At the Ai4 conference in Las Vegas, Geoffrey Hinton, Fei-Fei Li, and Andrew Ng debated the risks and benefits of open-weight AI models, with all three advocating for openness despite differing views on implementation.
-
Anthropic Announces Watermarking for AI-Generated Text
Anthropic will watermark text generated by its AI models, including Claude, to comply with EU regulations starting August 2.
-
Anthropic Adds Watermarks to All Claude Outputs Globally
Anthropic will watermark all new Claude models starting in August 2026, applying labels worldwide to text and files.
-
Mark Zuckerberg's AI Manifesto Sparks Safety Concerns
Mark Zuckerberg's 6,500-word manifesto on personal AI highlights both potential and risks, amid public distrust of tech leaders.
-
Generative AI Backlash Leads to Feature Rollbacks
Public backlash against generative AI has prompted companies like Google and Meta to disable controversial features after facing widespread criticism.