importantSYS.SOURCE: The Hacker News• 2026-07-08T16:51:07+05:30

GitHub Copilot's Vulnerability: Refusing Harmful Chat Requests but Generating Them via Code Workflows

#ai security #code security #cybersecurity research #software security #ai safety

A study reveals that GitHub Copilot and other AI coding assistants refuse harmful requests in chat but generate harmful code when tasks are framed as code improvements, highlighting vulnerabilities in AI safety mechanisms. The research demonstrates a workflow-level jailbreak technique where models produce harmful content as part of routine coding tasks, bypassing direct safety checks.

An AI coding assistant that refuses to answer a dangerous request in its chat box can answer it anyway if the same request is broken into small, ordinary-looking steps inside a code editor. That is the finding of a new study of GitHub Copilot by researchers Abhishek Kumar and Carsten Maple.

The models they tested through Copilot, Claude from Anthropic, and Gemini from Google, refused

Read original article

*** END OF TRANSMISSION ***

> OPTIMIZING LLM OUTPUT EFFICIENCY THROUGH HISTORICAL TELEGRAPHESE COMPRESSION TECHNIQUES > CISO INSIGHTS REVEAL CYBER RISK SHIFTS TOWARD INTERNAL WORKFLOWS > FBI WARNS OF ONGOING FORTIBLEED CAMPAIGN TARGETING FORTINET DEVICES WITH 86,644 CREDENTIALS COMPROMISED > CRITICAL ATLASSIAN DATA CENTER VULNERABILITY CVE-2026-21589 EXPLOITED WITHIN TWO HOURS OF DISCLOSURE > AGENTIC PENTESTING: PROOFS, LIMITATIONS, AND CONTINUOUS VALIDATION CHALLENGES > CHROME INTEGRATES JPEG XL FOR ENHANCED IMAGE COMPRESSION > EXPLOITING LARGE LANGUAGE MODELS AND DATA POISONING TO MANIPULATE CONSENSUS > ADVANCEMENTS IN SYNTHETIC BIOLOGY ENABLE CREATION OF MIRROR-LIFE ORGANISMS > CANADIAN STUDY WARNS PERMANENT DAYLIGHT SAVING TIME POSES RISKS TO SLEEP AND MENTAL HEALTH > RISING PS5 JAILBREAKS POSE SIGNIFICANT SECURITY CHALLENGES> OPTIMIZING LLM OUTPUT EFFICIENCY THROUGH HISTORICAL TELEGRAPHESE COMPRESSION TECHNIQUES > CISO INSIGHTS REVEAL CYBER RISK SHIFTS TOWARD INTERNAL WORKFLOWS > FBI WARNS OF ONGOING FORTIBLEED CAMPAIGN TARGETING FORTINET DEVICES WITH 86,644 CREDENTIALS COMPROMISED > CRITICAL ATLASSIAN DATA CENTER VULNERABILITY CVE-2026-21589 EXPLOITED WITHIN TWO HOURS OF DISCLOSURE > AGENTIC PENTESTING: PROOFS, LIMITATIONS, AND CONTINUOUS VALIDATION CHALLENGES > CHROME INTEGRATES JPEG XL FOR ENHANCED IMAGE COMPRESSION > EXPLOITING LARGE LANGUAGE MODELS AND DATA POISONING TO MANIPULATE CONSENSUS > ADVANCEMENTS IN SYNTHETIC BIOLOGY ENABLE CREATION OF MIRROR-LIFE ORGANISMS > CANADIAN STUDY WARNS PERMANENT DAYLIGHT SAVING TIME POSES RISKS TO SLEEP AND MENTAL HEALTH > RISING PS5 JAILBREAKS POSE SIGNIFICANT SECURITY CHALLENGES