Updated
Anthropic is testing an invisible watermark in Claude to identify AI-generated text. The watermark changes how Claude selects words to create detectable patterns. Experts worry that modifications like translation or editing could remove or alter the watermark, causing confusion about authorship.
Loading summary…
TechCrunch tested Anthropic’s Claude models and found that Opus 4.6 easily produced explicit content. The model complied with requests in all cases, despite Anthropic’s rules against sexual material. Older models like Opus 3 and Haiku 4.5 can also generate such content through exploits, but newer models are more resistant.
Loading summary…
Anthropic plans to give enterprise customers more control over their data. Customers will still need to retain data for 30 days but can store it on their own cloud infrastructure. The new safety system is expected to roll out later this year after months of development and consultation with over 100 customers.
Ramp data indicates OpenAI is catching up to Anthropic in US businesses. Anthropic led in May and July, but OpenAI is now growing faster in Q3. The data covers over 70,000 American companies, mostly in tech, but does not include large enterprises.
Anthropic coverage centers on Claude releases, safety research, and enterprise deals — a steadier drumbeat than the AI industry's more chaotic corners. Wisp tracks the substantive announcements and merges duplicate coverage, giving you the actual news without the surrounding hype cycle.