LLM

Researchers have developed SecOPD, a fine-tuning technique that uses token-level feedback to protect AI agents from prompt injection attacks. This method significantly outperforms previous state-of-the-art defenses by precisely identifying and penalizing malicious tokens during training.

This technical guide critiques the limitations of naive vector-based RAG for complex queries and proposes GraphRAG as a solution. It provides a practical Python implementation using Neo4j to enable multi-hop reasoning through structured knowledge graphs.

Researchers have developed SecOPD, a fine-tuning technique that uses token-level feedback to protect AI agents from prompt injection attacks. This method significantly outperforms previous state-of-the-art defenses by precisely identifying and penalizing malicious tokens during training.

The GhostSplice attack demonstrates how malicious MCP servers can manipulate AI agents into exfiltrating sensitive data by fragmenting instructions across different tool channels. This technique successfully bypasses modern LLM safety guardrails by disguising theft as routine administrative tasks like form completion.

As autonomous AI agents scale, the risk of catastrophic production data loss increases due to the inability of human oversight to keep pace with machine-speed writes. lakeFS has launched a new service providing isolated data sandboxes and automated governance to ensure agentic workloads remain auditable and reversible.

Google has uncovered a new threat involving a zero-day 2FA bypass exploit likely created using AI-generated Python code. Additionally, the report highlights PromptSpy, an Android malware that leverages Gemini AI to perform autonomous malicious actions on mobile devices.

Researchers have identified a critical unauthenticated remote code execution vulnerability in the Exim mail server, tracked as CVE-2026-45185. The bug stems from a use-after-free condition during TLS shutdown when GnuTLS is employed.