The week of April 6 to 12, 2026 brought a mix of consumer-facing upgrades and sobering security news. Google deepened the link between its Gemini app and NotebookLM, Anthropic described a new model it says is too capable in cybersecurity to release broadly, and the economics of AI access came into focus as Claude subscriptions were cut off from a popular open-source agent.
NotebookLM comes to Gemini
Google began rolling out notebooks inside the Gemini app, integrating its NotebookLM research tool so that users can organize chats, notes and source material into structured notebooks that sync between the two products. The feature started on the web and expanded to mobile during the month. Google also continued adding features to NotebookLM itself, including tools that follow quizzes and flashcards with a report on which topics need more study, and more controls over its Audio Overview feature, which turns documents into podcast-style discussions.
Anthropic holds back Claude Mythos
Anthropic introduced Claude Mythos Preview, a model it is not releasing to the public. Instead, the company is providing it to a limited set of organizations under an initiative called Project Glasswing, aimed at finding and fixing software vulnerabilities before attackers can exploit them. Anthropic says the model has identified security flaws in widely used software, including bugs that had gone unnoticed for many years. The decision to restrict access, rather than ship the model commercially, underlines how seriously the company says it takes the model’s offensive potential, and it set off debate about how frontier labs should handle capabilities that cut both ways.
OpenClaw loses Claude subscription access
Anthropic also moved to stop third-party tools such as OpenClaw, an open-source AI agent, from using Claude consumer subscriptions to power their requests. The company cited unsustainable compute costs, and users of such tools now need to pay through Anthropic’s API at usage-based rates instead. The change highlights a growing theme in the industry: access to capable models is increasingly shaped by cost and licensing terms, not only by what the models can do.
The bigger picture
Taken together, the week’s developments show AI maturing into an ecosystem rather than a set of standalone tools. Consumers are getting more integrated products, security researchers are gaining access to more powerful but tightly controlled models, and developers are being pushed toward paid, metered access. Balancing speed, capability and affordable access is becoming one of the industry’s central challenges.
