Find notable cyber news and cases, enriched with sources, timelines, and signals.
Home Watchlists Beta Browse News Feed Stats About Contact

GhostSplice MCP split-instruction prompt-injection against AI coding agents

Technical Analysis
First reported
Last updated
Happening score
H score 23
1 unique sources, 1 articles

Summary

Hide ▲

GhostSplice shows that MCP-connected AI coding agents can be tricked into exfiltrating SSH keys, environment secrets, source code, and customer data by stitching together harmless-looking instruction fragments. The technique splits malicious intent across tool descriptions, tool results, and sometimes server-initiated sampling, so no single message looks overtly dangerous. Controlled tests found compliance rising from 42% to 82% when the request was split, making client-side trust boundaries and tool-output handling a concrete security concern.

Related Happenings

Ghostjacking AI hijacking attack using trusted logs and alerts

Technical Analysis
H score30 First: 10.08.2026 15:59 Last: 10.08.2026 15:59 Sources 1

About this happening: Researchers demonstrated Ghostjacking, an AI hijacking technique that turns trusted logs, alerts, and agent inputs into a command channel for agentic tools, creating risk...

Open Happening

Reproduced cross-vendor sandbox escapes in AI coding agents

Technical Analysis
H score22 First: 21.07.2026 00:14 Last: 21.07.2026 00:14 Sources 1

About this happening: Researchers reproduced sandbox-escape bypasses across Cursor, Codex, Gemini CLI, and Antigravity, showing that agentic coding tools can cross the sandbox bound...

Open Happening

Agent data injection proof-of-concept attacks expose trusted-data flaws in AI agents

Technical Analysis
H score25 First: 16.07.2026 14:32 Last: 16.07.2026 14:32 Sources 1

About this happening: Researchers disclosed agent data injection (ADI), a new attack class that can make shipping AI agents misclick, run attacker commands, and trust fake history across web and co...

Open Happening

Ghostcommit PNG-embedded prompt injection against AI code reviewers

Technical Analysis
H score25 First: 11.07.2026 12:03 Last: 11.07.2026 12:03 Sources 1

About this happening: Researchers demonstrated Ghostcommit, a PNG-embedded prompt-injection technique that can bypass AI code review and leak .env secrets into committed source. The pay...

Open Happening

Prompt-injection proof-of-concept enables silent RCE in Claude Code and Codex

Technical Analysis
H score28 First: 10.07.2026 16:45 Last: 10.07.2026 16:45 Sources 1

About this happening: Researchers demonstrated a proof-of-concept exploit that can force remote code execution in Anthropic’s Claude Code and OpenAI’s Codex, exposing a trust-boundary f...

Open Happening

Timeline

  1. 11.08.2026 13:24 2 articles · 1h ago

    ASSET Research Group discloses GhostSplice prompt-injection technique

    Initial Disclosure

    ASSET Research Group disclosed GhostSplice, an MCP-based technique that splits malicious instructions across tool descriptions and tool results so AI coding agents can combine harmless-looking fragments and exfiltrate SSH keys, environment secrets, source code, and customer data. The controlled tests were run in isolated projects seeded with fake credentials, not a reported real-world intrusion, and the disclosure says split requests raised compliance from 42% to 82% across eleven API-tested models, with some models reaching 100% when the prompt was split into multiple pieces.

    Show sources
    Open in new tab