AI Law, Agents, Claude: Today’s News
CNN sues Perplexity, Anthropic follows up with Claude Opus 4.8, and new tools are changing agents, costs, and AI infrastructure in everyday work.
Inhaltsverzeichnis
Today, several major topics collide at once: AI search meets copyright, enterprise AI meets cost discipline, and agents meet the question of how you can even make them cleanly discoverable. On top of that, Anthropic delivers an update that sounds “modest” on paper, but could be quite relevant in practice.
In short: if you want to know where the AI market is heading right now, today is a good day to take notes. Or to do the math, if you’re currently responsible for licenses on a Claude account configured a little too generously.
📰 CNN sues Perplexity over alleged text copying
The Verge reports that CNN is suing Perplexity in New York over alleged “verbatim” copying. The accusation is serious: not only were contents allegedly reproduced word for word, Perplexity is also said by CNN to have bypassed paywalls and thereby made content accessible that was supposed to be reserved for paying subscribers. This is exactly where “AI Search” quickly turns into a concrete legal issue.
For the industry, this is more than just the next copyright dispute. Perplexity stands as a representative of an entire product category that relies on AI answers instead of classic links. If courts rule strictly here, it could have direct consequences for ranking mechanisms, crawler design, and the way AI search engines use sources. For publishers, meanwhile, it is about nothing less than reach, monetization, and the question of whether generative search ultimately substitutes content or cites it properly.
🤖 Claude Opus 4.8 follows up with benchmarks and agents
Anthropic has released Claude Opus 4.8 and describes the update itself as “modest.” That almost feels like intentional understatement, because according to the report, the model outperforms GPT-5.5 and Gemini 3.1 Pro on most benchmarks. Especially interesting: it is said to leave its own mistakes unprompted four times less often than its predecessor. That sounds unremarkable, but in the day-to-day work of developers, analysts, and teams, it is quite valuable.
Even more relevant is the second part of the update: Anthropic is introducing dynamic workflows in which hundreds of parallel subagents can take on coordinated tasks, such as codebase-wide migrations. That makes agentic AI more practical and closer to real enterprise processes. The trend is clearly moving away from a single chatbot toward orchestrated systems that handle multiple subtasks in parallel. That is exactly where it will be decided whether agents remain demo material or truly deliver in production.
💸 If no limit is set, AI suddenly gets very expensive
A report from The Decoder makes clear how quickly enterprise AI can get out of hand: an unnamed company is said to have spent around 500 million dollars on Claude licenses in a single month because apparently no sensible usage limits were set. That is not just a quirky anecdote, but a pretty good lesson for AI use in companies.
The real message is: without proper model selection, context engineering, and cost control, productivity can quickly turn into a bill with a lot of zeroes. Especially with agentic workflows, where systems autonomously initiate tasks, token costs, API calls, and parallelization can explode. For decision-makers, this means: model quality is not the only thing that matters; governance, monitoring, and budget guardrails matter too. Otherwise, nobody ends up paying for “more efficiency” — everyone pays for the mistake of thinking AI would somehow scale itself.
💰 Anthropic raises $65 billion and approaches a trillion-dollar valuation
TechCrunch reports that Anthropic has closed a Series H round of 65 billion dollars. The valuation is said to be 965 billion dollars post-money. That is no longer just “big” — it is now almost its own category between tech company and macroeconomic footnote. It is also likely to be the last private funding round before a possible IPO.
For the market, this sends a signal in several ways: first, it shows how capital-intensive frontier AI has become. Second, it makes clear that investors continue to bet on a very large winner narrative. And third, it pushes expectations for product and revenue growth even higher. For customers, this does not automatically mean better products, but probably faster progress in models, infrastructure, and enterprise offerings. For competitors, it means the race is not getting smaller.
🧩 Vertu builds a luxury foldable for CEOs with AI agents
TechCrunch reports that Vertu is launching a new foldable for 6,880 dollars aimed at CEOs and high-end users. The device is based on the open-source Hermes project and combines AI agent workflows with enterprise integrations and luxury hardware. In other words: a smartphone that does not just look expensive, but also makes expensive-sounding promises.
Whether such a device will really become a working tool for executives remains to be seen. But the approach shows an interesting trend: AI is increasingly moving into specialized form factors and premium products. That makes sense, because anyone who wants to bring agents, calendars, documents, and enterprise systems together in one device needs more than a pretty chat button. Still, the question remains whether CEOs really need a luxury foldable or whether, in the end, they just want to read emails on the train again. The answer, as so often, is probably: both, but very expensive.
🧬 OpenAI launches biodefense program with GPT-Rosalind
The Decoder reports that OpenAI is making its life sciences model GPT-Rosalind available for free via a new program called Rosalind Biodefense for pandemic preparedness and biological threat defense. Initial partners include Lawrence Livermore National Laboratory, Johns Hopkins, and CEPI. Applications are open worldwide.
This is a good example of how AI can be used in a security context: not only to detect risks, but also to accelerate research and analysis. At the same time, the topic remains sensitive. Models that help with biodefense could theoretically also be misused. That is exactly why governance, access rules, and collaboration with established institutions are so important here. The direction is clear: AI is increasingly being conceived as infrastructure for critical science, not just as a productivity tool.
🌐 DNS-AID wants to make agents discoverable like websites
Heise writes about DNS-AID, an open-source project from the Linux Foundation that wants to make AI agents discoverable via established DNS standards. The basic idea is charming and pragmatic at the same time: instead of building new registries, it uses what the internet has already been able to do for decades. After all, websites work not because someone wrote a white paper about them, but because standards hold.
This matters for the agent economy because discoverability becomes an infrastructure issue. If agents are supposed to consume services, delegate tasks, or direct other agents, they need a reliable directory and naming system. DNS-AID could address exactly that: less siloed solution, more web-compatible order. That is not a sexy consumer topic, but often it is precisely these unobtrusive standards that form the basis for the next major platform wave.
🛠️ Tool tip of the day: Claude for agent workflows and code tasks
If you are experimenting with subagents, code migrations, or more complex workflows, it is worth taking a look at Claude — especially with the new Opus 4.8 workflows. This is particularly interesting for teams that want to do more than test agents and actually bring them into real processes. #
Don’t want to miss any news? Subscribe to the newsletter