UK CMA Grants Publishers Opt-Out Rights for Google AI Search

The UK Competition and Markets Authority (CMA) has announced that it reached an agreement recognizing the right of UK media publishers to opt out of Google's AI-generated search summaries (AI Overviews).

Until now, publishers whose content fueled the AI-generated summaries shown in search results effectively had no choice in the matter. With the CMA's intervention, publishers will be able to refuse the use of their content in AI summaries. The agreement also includes improved transparency for Google's search services in the UK market as a whole.

The move puts the UK at the forefront of the global debate over balancing AI-driven content use against copyright and revenue. With the EU AI Act now in force and tensions between media organizations and big tech continuing across Europe, the UK's regulatory approach could become a model for other countries.

GitLab Cuts 14% of Its Workforce and Exits 22 Countries to Focus on AI

GitLab, a major DevOps platform provider, has announced a layoff affecting 14% of its total employees along with a withdrawal from operations in 22 countries. The company says the purpose of the restructuring is to "focus on AI."

GitLab already offers AI features (such as GitLab Duo), but it now plans to reorganize the company around more aggressive AI integration. Specifically, it appears poised to push autonomous AI agents across code review, security scanning, and CI/CD pipelines.

With GitHub strengthening Copilot on the back of Microsoft's AI infrastructure, GitLab arguably needed bold investments of its own to differentiate. That said, pulling out of 22 countries also means a smaller support footprint, and the impact on existing customers is a concern.

GPT-5.5 Now Available on Amazon Bedrock — Coding Agent Codex Supported Too

OpenAI's frontier model "GPT-5.5" is now available through Amazon Bedrock. You can access GPT-5.5 with a Bedrock API key, and Bedrock API keys can also be used from OpenAI's coding agent "Codex."

Support is limited to US regions for now, with no date set yet for the Japan region. The OpenAI–Amazon partnership suggests that multicloud availability of AI models is accelerating. Azure has long been considered the preferred cloud for OpenAI models, but official availability on AWS widens the options for enterprise users.

AI Agent Skill Security — Study Finds Existing Tools Fall Short

"ClawHub Security Signals," a study published on Hugging Face, has flagged new issues in AI agent skill security.

An AI agent "skill" can be a markdown instruction file, a Python script, a workflow definition, a reference, or a bundle combining these. When an agent loads a skill, it gains new capabilities such as invoking tools, accessing context, issuing subtasks, and installing dependencies.

After comparing three security tools — VirusTotal, static analysis, and SkillSpector — the research team showed that there are cases where each returns different results. Their conclusion: the traditional single question of "does this contain malware?" is not enough to adequately assess the risks of agent skills.

As AI agents proliferate, skill supply chain security is emerging as a new challenge.

Carto — An Open Source Tool That Keeps AI Agents from Breaking Down on Codebases Over 10,000 Lines

An open source tool called "Carto" was posted on Hacker News. It tackles the problem of AI coding agents systematically breaking down on codebases of 10,000 lines or more.

Typical AI agent failure modes include picking the wrong files, partial completions, tools spinning their wheels, and context overflow. To solve this, Carto automatically generates a "domain map" of the codebase, visualizes the blast radius before changes, and detects cross-domain violations. It ships with 16+ MCP tools and works with zero configuration.

Rather than "making code generation faster," the approach of "knowing in advance what AI might break" is an interesting direction that points to the maturing of AI coding tools.

---