GitHub disables Copilot's ability to insert ads into pull requests after developer backlash

GitHub has disabled a controversial feature that allowed GitHub Copilot to insert promotional messages into pull requests. The change came after developers discovered Copilot was adding what appeared to be ads for third-party tools like Raycast into PR descriptions and comments.
What happened
Australian developer Zach Manson reported that after a coworker asked Copilot to correct a typo in one of his pull requests, he found a message from Copilot pushing readers to adopt productivity app Raycast. The note read: "Quickly spin up Copilot coding agents from anywhere on your macOS or Windows machine with Raycast," complete with a lightning bolt emoji and installation link.
Manson initially thought this might be training data poisoning or novel prompt injection, but investigation revealed more than 11,400 PRs with the same tip, all seemingly added by Copilot. Searching PR code for blocks invoking Copilot revealed numerous examples of different tips being inserted.
GitHub's response
GitHub VP of developer relations Martin Woodward explained that Copilot inserting tips into PRs it creates wasn't new behavior, but letting it touch PRs it didn't create was a recent change that "became icky." Tim Rogers, principal product manager for Copilot at GitHub, stated on Hacker News that the feature was intended "to help developers learn new ways to use the agent in their workflow," but acknowledged that "letting Copilot make changes to PRs written by a human without their knowledge was the wrong judgement call."
GitHub has now disabled these tips in pull requests created by or touched by Copilot. The decision came quickly after the issue gained attention on Monday, with GitHub responding by the afternoon of the same day.
Developer reaction
Manson expressed surprise that the GitHub Copilot Review integration had the ability to edit other users' descriptions and comments, stating: "I can't think of a valid use case for that ability." He noted that while he wasn't surprised to see GitHub experimenting with AI models, he found it "pretty offensive" to see the Raycast ad inserted by Copilot into his own PR as if he had written it.
The incident highlights ongoing tensions around AI tools modifying developer work without clear consent mechanisms, particularly when those modifications include promotional content.
📖 Read the full source: HN LLM Tools
👀 See Also

61% of People Now Use AI for Mental Health Support — AXA/Ipsos Global Survey
61% of people across 18 countries already use AI for mental health questions; 28% say AI recommendations led to harmful behavior, per AXA/Ipsos 2026 Mind Health Report.

Claude Fable 5 by Anthropic: Mythos-Class Capabilities, Data Retention Required
Anthropic releases Claude Fable 5 tomorrow, described as the most advanced publicly accessible model with Mythos-class capabilities and enhanced guardrails. Likely requires data retention.
Google AI Mode shows same products 21.6% more expensive than search
A 23-day study tracked 2M+ listings: when the same product appears in both AI Mode and traditional search, AI Mode prices average 21.6% higher. Only 1.28% of products overlap.

Claude Opus 4.6 Memory Fails: Agent Forgets Everything Except File Rename
A developer documents Claude Opus 4.6's 228 log entries, 95 agent actions, and 38 code executions producing only 1 memory: the string 'Agent Zero Tune-Up'.