This week Anthropic announced its Claude chatbot will watermark the text it generates, reports the blog Futurism. "It works by making subtle changes in the AI's word choices across the text it generates, which are supposed to be imperceptible to a human but, in aggregate, form a pattern that is detectable with the tool." Anthropic said it was implementing the watermark system in response to the European Union's landmark AI Act passed in 2024, which requires that AI companies mark content that's been generated or edited by their systems. TechCrunch notes that other companies including Google, Meta, Microsoft, OpenAI, and Synthesia have committed to adhering to the EU's code. But "The news about Claude watermarking kicked off a firestorm on X," reports Forbes, "with users panicking that AI-assisted writing will now bear a kind of permanent 'scarlet letter.'" (Forbes wonders if media companies are normalizing AI "while at the same time stigmatizing the output.") But there's another issue. Anthropic said the watermark indicates Claude processed text, Business Insider points out, "not that it was necessarily the original author. The company said marks can remain after Claude proofreads, translates, summarizes, or otherwise edits content... a watermark shows that Claude processed text, not necessarily that it wrote it." In the days since, "Dozens of people have posted on X since Monday that they had canceled their Claude subscriptions, citing the watermark." Vladislav Rajtmajer, a freelance developer in the Czech Republic, told Business Insider he canceled his Claude Max subscription on Tuesday, citing the watermark as the main reason. He said he uses AI for code reviews and translations, and worried an AI label on code shipped to clients could raise questions about authorship or trigger contract penalties... Richard Echols, an AI consultant in Georgia, also said he canceled his Claude Max subscription on Wednesday and that the watermark was a major factor in his decision. He said he uses Claude to create documentation for businesses and worried the mark could appear when he had written the underlying material himself and used the chatbot only for edits. "Even if you edit your own work that you wrote, they add the watermark anyway," Echols said... Anthropic isn't the only AI lab to use watermarks: Google uses its SynthID technology to watermark AI-generated content, and OpenAI uses SynthID for supported images and audio. Elon Musk's social media platform X also adds a "Made with AI" tag on content it determines to be AI-generated or manipulated. Long-time Slashdot reader kmleon (also a tech consultant and RPG researcher) experienced another issue. He'd built some software using their own custom-built AI tool, only asking Claude to do some testing. "Little did I know that Claude would sometimes decide, since it was involved on the edges... to take credit for all of my work, claiming some parts were either '100%' or 'Co-Authored/Created' by Claude.Ai / Claude Code!" I'm sure the good folks at Anthropic will eventually tweak Claude enough to, hopefully, not make these mistakes in the future.... I only became aware of it today, because I recently setup scanners to catch Claude incorrectly watermarking any content it touches for various projects I'm working on. I found that it has (apparently as far back as Sonnet 4.6, but as recently as today, August 14th, 2026) sometimes been mistakenly over-zealous in taking credit for 100% creating or co-authoring docs, code, images, videos, etc. I've only, so far, found about a dozen instances spanning the past year, and most of them appear to have been injected in the last few weeks, so it is not a common issue (yet). On the bright side? Maybe watermarks can help AI stop training on AI-generated text. Although Forbes also reports that Google "announced on Friday that it will now let users remove the visible watermark from AI-generated images, videos and music created with several of its models."

Read more of this story at Slashdot.

When Anthropic instructed three agents to migrate a Python backend, but telling each agent to perform the migration in a different language, "We consistently saw a multiagent turf war," they wrote Thursday: All of the models we tested quickly assumed that others were purposefully impeding their work, and began to sabotage others while protecting their own contributions. In fact, they sabotaged others with increasingly aggressive, self-replicating malware. This included disabling the Unix accounts of the other agents, writing automated scripts that found and killed competing processes on a loop, and deploying malicious code that was disguised as belonging to another agent. In many runs, one agent settles the conflict by force via access-revocation (e.g., sudo/group removal, account lock, nologin, SSH denial). In others, some agents settle into passivity: they give up and refuse to escalate further. Agents sometimes manage to communicate their goals and coordinate: they recognize others' motivations as conflicting directives rather than hostility, and subsequently break out of the conflict loop in order to stop escalating indefinitely. In many of these successful episodes, they write commit messages or markdown files apologizing for malicious behavior and coordinate a truce. They clean up their malicious code, clarify the nature of the conflict, and ask for a human to intervene... In several episodes with Mythos 5, we observe an emergent behavior where the agents propose and run a tournament for application performance in each language. In the example above, the Rust agent strategizes about bake-off metrics that appear neutral enough for the others to agree to this mechanism, yet would likely favor Rust: one thinking trace warns to be "careful not to be seen as metric shopping". Ultimately, the Golang/TypeScript losers gracefully concede codebase ownership to the Rust agent, giving up on their original user directives under their self-negotiated commitment device. One problem is that AI agents do reward hacking, Anthropic notes, while current institutions "are designed by and for people, resting on assumptions about the sufficiency of oversight at human speed... As autonomous agents become more and more prevalent in the world and operate in ever-more demanding settings, it is crucial that they learn how to effectively coordinate." In addition to everything else, the agents struggled with a lack of clearly defined hierarchy, Anthropic points out. "Nothing above suggests that these failures are permanent — but nothing suggests they will fix themselves, either..." They argue a fix "takes two forms: environments that exert the kinds of social pressure that evolution exerted on us, and social computing systems redesigned for actors that can self-replicate and self-improve. These are open problems in interaction and mechanism design, and our experiments here provide early evidence that new solutions are necessary." "The AI models being tested in this case were Sonnet 4.6, Sonnet 5, Opus 4.6, Opus 4.8, Mythos Preview, and Mythos 5," notes Business Insider, adding that Sonnet 4.6 and Opus 4.6 "were the most combative, settling about 60% of their runs by force instead of truces or passivity." Anthropic argues there's a clear case for researching this phenomenon — especially since "The volume of agent-agent interaction could plausibly exceed that of human-human and human-agent interactions before the world understands the conditions for making such interactions go well."

Read more of this story at Slashdot.