AI Daily Briefing — 2026-06-24

Anthropic has launched Claude Tag, a new research preview feature that embeds an always-on AI assistant directly into Slack. Designed as a persistent teammate, Claude Tag can be mentioned in channels to provide insights, assign tasks, and even proactively jump into conversations. The tool builds on existing Slack integrations but adds a layer of continuous context and memory, allowing the AI to learn from ongoing work and surface relevant information across the organization—provided it has permission to access other channels.

With Claude Tag, every user in a Slack channel shares access to a single Claude identity. This means anyone can see what the AI has been working on and continue conversations where others left off. System administrators control which tools, information, and channels Claude can access, ensuring that a Claude set up for legal work, for example, cannot bleed into engineering discussions. When assigned a task, Claude Tag breaks it into stages and works through them using available tools, posting updates directly in the thread.

Claude Tag also features an ambient mode that allows it to proactively flag updates, follow up on forgotten threads, and share information from across the organization. Anthropic says this makes the AI feel like a real colleague, capable of producing work in public view with deeper context and understanding. This push for persistent enterprise context mirrors efforts by Microsoft, Snowflake, Databricks, and Glean, all of which are building systems that tap into organizational knowledge to power AI agents.

OpenAI has unveiled a new program called “Patch the Planet,” aimed at bolstering cybersecurity within the open source community. The initiative, announced on Monday, partners OpenAI with security firm Trail of Bits to help maintainers identify and fix vulnerabilities in their projects. The name is a nod to the 1995 film “Hackers,” but the mission is serious: reducing the burden on overworked open source developers.

Under the program, Trail of Bits engineers will work directly with project maintainers to review code and develop patches, supported by OpenAI’s security tools like Codex Security. OpenAI stated that the goal is to streamline the process, ensuring that security issues are triaged before reaching maintainers and that reusable workflows are created for ongoing protection. The effort addresses a critical gap, as open source software often suffers from security flaws due to limited resources and oversight.

The initiative comes amid growing concerns about AI-powered tools, such as Anthropic’s Mythos, which can automatically identify and exploit bugs. OpenAI’s approach flips this dynamic by using AI defensively, helping the open source community protect itself. While questions remain about long-term scalability, the program represents a significant step toward securing the digital infrastructure that underpins much of the commercial software industry.

Deep beneath the North Sea, engineers are completing the world's longest and deepest subsea road tunnel. The Rogfast project in Norway stretches 16.6 miles and reaches 1,280 feet below sea level. Journalist Niall Firth recently visited the site, describing the eerie pressure of millions of tons of seawater overhead. The tunnel represents a triumph of ambition in an era when large-scale infrastructure projects often face delays and skepticism.

Meanwhile, the AI boom is straining the electric grid, but a solution may lie in smarter power management. Rather than building new power plants, data centers can reduce their energy draw during peak demand. New software tools are making this "flexibility" faster and more precise, helping grid operators balance loads without waiting for costly upgrades.

These stories highlight how engineering innovation continues to solve complex challenges. From record-breaking tunnels to adaptive energy systems, the drive to build and improve remains strong. The Rogfast tunnel proves that ambitious construction is still possible, while flexible data centers show how existing infrastructure can be optimized for the AI era.

ASML's latest lithography machine, roughly the size of a double-decker bus and weighing over 150 tons, costs $400 million. It uses extreme-ultraviolet light, generated by firing lasers at molten tin droplets tens of thousands of times per second, to etch the tiny features needed for the world's most powerful chips. As demand surges in the AI era, ASML now controls about 90% of the global chip-lithography market, a dominance that has sparked unease among governments and attracted new competitors aiming to challenge its position.

Meanwhile, Anthropic recently announced it had built an AI model called Mythos, which it deemed a potential cybersecurity risk. The company released a safer version, Fable, but days later the U.S. government imposed export controls on both models, prompting Anthropic to revoke access entirely. Critics note this intervention targeted a coding model rather than a bioweapon or rogue AI, suggesting the response is more reactive than strategic.

Experts are watching three key aspects of this standoff: the implications for AI safety policy, the precedent set by government intervention in model releases, and how this might shape future regulation. The clash highlights the growing tension between innovation and security in the AI landscape, as reported by MIT Technology Review AI.

Nvidia has unveiled a warm-water cooling system it claims can nearly eliminate water consumption inside AI data centers. The closed-loop design circulates coolant at 113°F through server racks, then releases heat through passive radiators, removing the need for evaporative cooling or fans. An Nvidia executive recently told Axios that this largely solves the water challenge for data centers. However, the company's definition of water use stops at the facility walls, leaving a much larger footprint unaddressed.

The real issue lies beyond the data center. Power plants that supply electricity to these facilities consume vast amounts of water—natural gas uses 1.17 liters per kilowatt-hour, coal 2.2 liters, and hydropower loses 6.8 liters per kilowatt-hour through reservoir evaporation. Since fossil fuels still generate about half of data center power, according to the IEA, the total water footprint can double or triple when accounting for energy production and chip manufacturing. Nvidia’s solution only tackles roughly a quarter to a third of that total.

While the new system is a step forward—it also boosts efficiency and reduces noise—it doesn't address AI's broader water problem. Without a shift toward solar or wind, which use negligible water, data centers will continue to drain resources indirectly. TechCrunch AI has reached out to Nvidia for further clarification on these off-site impacts.