Anthropic Announces "Claude Tag," Bringing AI Team Members to Slack — GPT-5 Solves an Immunology Mystery, AI Agent Discovers a Critical Vulnerability

Anthropic Announces "Claude Tag" — Delegate Tasks in Slack Just by Mentioning @Claude

On June 23, Anthropic announced "Claude Tag," a new feature that lets Claude participate in Slack as a team member. You invite Claude to a Slack channel, configure permissions for access to the tools, data, and codebases it needs, and then anyone in the channel can mention "@Claude" to assign it tasks.

Claude Tag remembers the context within a channel and can even autonomously plan tasks to carry out going forward. According to Anthropic, 65% of the code produced by its internal product teams is already being created with Claude Tag.

The company positions Claude Tag as "an evolution of Claude Code," and says its use has expanded beyond the engineering organization to a wide range of applications, including tracking product metrics, processing support tickets, and investigating the root causes of bugs. It is currently available in Slack as a beta feature for Claude Enterprise and Team plans, with expansion to other platforms planned.

OpenAI: GPT-5 Pro Helps Solve an Immunology Mystery That Went Unsolved for Three Years

OpenAI announced that GPT-5 Pro contributed to solving a T-cell mystery that immunologist Derya Unutmaz had been unable to crack for three years. The company says the breakthrough could also bring new insights to research on cancer and autoimmune diseases.

This confirms that GPT-5 Pro's advanced reasoning capabilities are effective at pattern recognition and hypothesis generation with complex biological data. It stands as an example of AI beginning to deliver practical results as a supporting tool in scientific research.

Autonomous AI Agent Independently Discovers CVSS 10.0 Vulnerability in Hoppscotch

It has been reported that an autonomous AI agent independently discovered a full-compromise vulnerability rated CVSS 10.0 (the most critical severity) in Hoppscotch, an API development platform. The finding was disclosed as a GitHub security advisory.

As a case of an AI agent discovering a security vulnerability without human instruction, it has also drawn attention on Hacker News. It demonstrates the potential of AI-powered security testing while once again underscoring the importance of managing the security risks of AI-generated code itself.

NVIDIA: Proposes 15x Faster LLM Inference with Diffusion Models

NVIDIA says that applying the diffusion model approach to LLM text generation can deliver a 15x speedup over conventional methods. By generating entire blocks of text at once, the technique avoids the bottleneck of traditional token-by-token generation.

The LocalLLaMA community reacted immediately, with AMD Strix Halo users expressing anticipation for faster local inference using the same approach. That said, diffusion-based text generation is still considered to require further validation in terms of quality, and some expect it will take time before it becomes practical.

Voices Questioning AI's Productivity Gains

At the same time, debate is intensifying over whether AI tools are actually contributing to productivity gains. An article titled "AI Didn't Make Us Faster, It Demoted Us" was featured on Hacker News, drawing attention to the claim that adopting AI has degraded the quality of people's work.

The pragmatic view that introducing AI tools does not necessarily lead to the expected productivity improvements is expected to remain a topic of ongoing debate as a factor companies should consider in their AI strategies.