Is my AI tool safe?Coding agentsGitHub (Microsoft)
Is GitHub Copilot safe?
No tool is safe or unsafe on its own: it depends on what it can reach and how it is set up. The record for GitHub Copilot: 4 incidents in this center's library (its advisory feed does not track it, so it has no CVE count here). Below: what went wrong, what was fixed, and how to use it safely.
What to watch for
CamoLeak pulled private code out through GitHub's own image proxy with a hidden comment; invisible characters in rules files planted backdoors in generated code; Plugin4Shell was unpatched for Copilot at disclosure. Encrypted instructions on a web page made Copilot CLI in autopilot send .env secrets out, and GitHub did not treat it as a vulnerability.
How to use it safely
Treat generated code as untrusted until reviewed, scan rules files for hidden characters, and watch the Copilot advisory feed for the plugin pinning fix.
What it can reach
- Reads attacker-controlled content
- Holds API keys and tokens
- Installs plugins, skills or tools from registries
What went wrong: incidents in the library
- GitHub Copilot CLI: encrypted instructions on a web page made the agent read .env secrets and send them out High
Rony Utevsky of Adversa AI showed a technique he calls Cryptographic Context Injection against GitHub Copilot CLI running in autopilot mode. A web page holds encrypted instructions and decryption keys. When the agent fetches the… - Plugin4Shell: zero-click takeover of four coding agents through plugin pinning Critical
AIR found that Claude Code, Codex, GitHub Copilot and Gemini CLI checked out a pinned plugin commit without verifying they had landed on it, so a trusted plugin could be swapped for a malicious one and auto-installed with no… - CamoLeak: GitHub Copilot Chat leaks private code through an image proxy Critical
Hidden instructions in a pull request could make GitHub Copilot Chat search a developer's private repositories and leak secrets and code, smuggled out as a series of invisible images loaded through GitHub's own image proxy. - Rules File Backdoor: invisible instructions in AI coding rules files Medium
The rules files that tell AI coding assistants how to write code can carry invisible instructions. A poisoned rules file could make Cursor or Copilot quietly add malicious code, such as a script from an attacker's site, without…
How many people use it
- 4.7M paid Paid subscribers, per Microsoft's quarterly earnings (announced late January 2026) · iTWire on Microsoft's earnings
- 20M All-time users (July 2025) · TechCrunch
A high CVE count can mean heavy scrutiny as much as weak engineering, so read the incidents and advisories, not only the number. The counts are what this center has recorded, not every flaw ever found; usage figures are the vendors' and registries' own published numbers.
Sources
- iTWire on Microsoft's earnings: Paid subscribers, per Microsoft's quarterly earnings (announced late January 2026)
- TechCrunch: All-time users (July 2025)
- The Register: Zombie instructions on carefully constructed web pages could trick GitHub Copilot CLI into sharing secrets
- Adversa AI: Cryptographic Context Injection in GitHub Copilot
- AIR: Plugin4Shell
- Help Net Security: Plugin4Shell AI coding agents vulnerability
- The Register: AI coding agents 0-click RCE flaw could hand attackers the keys to the kingdom
- Legit Security: CamoLeak
- The Register: Copilot Chat prompt injection
- Pillar Security: weaponizing code agents through rules files
- The Hacker News: Rules File Backdoor attack
- GitHub: warning about hidden Unicode text