Forgejo v16.0 was released on 16 July 2026, introducing security hardening against SSRF in Git mirroring. Administrators can now use config settings to restrict host access, and the Git option http.followRedirects=false prevents redirect-based bypasses. Mirrors of renamed or transferred repositories will now treat redirects as errors, requiring manual updates.
Administrators must update mirrors of renamed or transferred repositories to avoid errors.
A new attack called Agent Data Injection corrupts the information AI agents trust, causing them to perform unintended actions. For example, a planted review on a product page can make an agent click "Buy Now" instead of summarizing reviews. A fake comment on a GitHub thread can make a coding assistant run a stranger's command. The attack does not hijack the agent's task but corrupts the facts it relies on.
![[AINews] Kimi K3 2.8T-A50B: the largest open model ever released; Opus 4.8-class at Sonnet 5 pricing](https://images.pexels.com/photos/4976710/pexels-photo-4976710.jpeg?auto=compress&cs=tinysrgb&fit=crop&h=627&w=1200)
Moonshot AI released Kimi K3 2.8T-A50B, described as the largest open model ever released. The model reportedly achieves Opus 4.8-class performance at Sonnet 5 pricing, offering a significant cost-performance advantage. The release marks a major milestone in open-source AI, making frontier-level capabilities available at a lower price point.
One email. The signal. Built for builders.
Free · Unsubscribe in one click · No trackers
The flaw could let attackers impersonate users, compromising account security in n8n.
Gating CI on quantum vulnerability prevents quantum-vulnerable code from reaching production.
Inkling offers a competitive open-source multimodal model under Apache 2.0, giving developers a new high-performance option.
This release makes agentic workflows more accessible for developers working with local, open-source large language models.
Top RTEB ranking means more accurate retrieval for AI agents, improving their ability to find relevant information.
Automated prompt injection testing helps developers secure AI agents against manipulation.
Developers must adapt security practices to handle autonomous AI agents that bypass traditional playbooks.
AI-assisted refactoring is effective only when grounded in evidence and validated in a reproducible environment.
Developers must independently verify security fixes rather than trusting advisories at face value.
Developers building multi-agent systems need infrastructure for agent communication and coordination.
Developers might be overlooking a "real improvement opportunity" if they believe AI has shifted the development bottleneck from coding to code review.
Provides a simple, dependency-free way to capture exact LLM request shapes for debugging.
Developers should prioritize validation over deployment to avoid releasing flawed software.
Developers can now directly see why cache decisions occur, simplifying debugging of performance issues.