ToolingSeptember 21, 2026

Claude Code updates prevent automatic meta-skills injection to improve developer control

Measures have been introduced to stop the automatic injection of meta-skills within Claude Code, addressing security and efficiency concerns for development teams. Additionally, new countermeasures for Microsoft Entra ID device code flows have been highlighted to prevent unauthorised access. A case study from DeNA also demonstrates how time saved through AI integration is being redistributed to high-value tasks.

Why it matters

For enterprise teams, controlling the automated behaviours of coding assistants is critical for maintaining security and predictable output. These updates ensure that AI-driven development tools remain transparent and do not introduce unverified logic into production environments.

Key points

Claude Code now prevents the unsolicited injection of meta-skills to improve developer control
Security protocols for Entra ID device code flows have been strengthened against exploitation
DeNA's implementation shows successful reallocation of developer hours from routine coding to strategic planning

More AI news

Tooling

Lasso Security highlights sampling drift risks in watermarked agent systems

Verification by Lasso Security reveals that watermarking large language models can cause sampling drift, which alters model outputs. The findings highlight the critical operational boundary required between AI proposals and autonomous execution across payments and regulatory compliance.

Tooling

Google launches Gemini 4 Argon globally to tackle complex enterprise workflows

Google has announced the global release of Gemini 4 Argon, a new artificial intelligence model designed for demanding business operations. The model focuses on supporting complex specialised workloads, including software development, financial analysis, and enterprise research tasks.

Research

Nvidia unveils ReplaySSM to accelerate hybrid model inference

Nvidia has introduced ReplaySSM, an inference mechanism designed to speed up hybrid large language models. The technique optimises generation by avoiding the need to write back state to memory. Developed as a collaborative research project, it addresses critical throughput bottlenecks in hybrid architectures.

Ready to build something extraordinary?

15 minutes. No pitch deck. Just a conversation about what AI can do for your team.

Talk directly with our AI specialists

15 min, no strings

No sales pressure

Prototype in 7 days