Skip to content
konsole

Resources · Threat intelligence

Threat intelligence

Researched reports on how AI systems are actually being attacked: prompt injection against deployed copilots and agents, attacks through the AI toolchain, and the ways model usage leaks data and money. Dated, sourced, and mapped to the controls that would have stopped them.

Sources

Every claim cited

Period

2025 – 2026

Watch

Ongoing

EchoLeak · zero-click exfiltration from Microsoft 365 Copilot · CVE-2025-32711nx “s1ngularity” · npm worm runs Claude Code and Gemini CLI to hunt secretsShadowLeak · ChatGPT Deep Research exfiltrates Gmail from OpenAI’s own cloudLLMjacking · stolen model keys run up to US$1M per victim (Google, Nov 2026)Clinejection · first in-the-wild prompt injection ends as an npm supply-chain compromiseGrok · 370,000 shared conversations indexed by GoogleForcedLeak and SalesBleed · Salesforce Agentforce exfiltrated through web-to-lead formspostmark-mcp · first malicious MCP server blind-copies every email it sendsDeepSeek · open ClickHouse database with a million lines of chat logs and keysGTG-1002 · espionage campaign with 80–90% of the work done by an AI agentShai-Hulud 2.0 · 796 npm packages, wipes the home directory if it finds no token175,000 Ollama servers answering on the public internet
21
prompt-injection incidents on record

May 2025 to Nov 2026

174
primary and press sources cited

every figure attributed

3
CVEs out of 21 incidents

the rest never reached a feed

6.2k
words of researched analysis

3 reports, updated monthly

  1. THREAT INTELLIGENCETRACKINGData leakage and runaway spendChat histories and keys left on the open internetShared conversations indexed by search enginesStolen model credentials and six-figure loopsDATED · SOURCED · MAPPED TO CONTROLS
    15 min readThreat reportData leakageCost

    Where AI usage leaks data and money: exposed infrastructure, shared chats, shadow AI, LLMjacking and runaway spend, 2024 to 2026

    A sourced report on the non-exotic ways organisations lose data and budget through LLM use: databases left open, chat links indexed by Google, employees pasting source code into personal accounts, stolen model credentials billed to the victim, and agents that nobody capped. With the figures, the regulators’ responses and the controls that address each.

    Read the report
  2. THREAT INTELLIGENCETRACKINGAttacks through the AI toolchainPackages that weaponise installed coding agentsBackdoored MCP servers and IDE extensionsIntrusions run end to end by an AI agentDATED · SOURCED · MAPPED TO CONTROLS
    17 min readThreat reportSupply chainAgents

    Attacks through the AI toolchain: worms that weaponise coding agents, poisoned MCP servers, and intrusions run by AI, 2025 to 2026

    A sourced report on the AI software supply chain under attack: npm worms that turn installed Claude Code, Gemini CLI and Amazon Q into secret hunters, backdoored MCP servers and IDE extensions, remote-code-execution flaws in Cursor, Copilot, Gemini CLI and Claude Code, and the vendor threat reports that document AI agents running intrusions end to end.

    Read the report
  3. THREAT INTELLIGENCETRACKINGPrompt injection in the wildZero-click exfiltration from enterprise copilotsCalendar invites that control smart homesBrowser agents hijacked by the pages they readDATED · SOURCED · MAPPED TO CONTROLS
    16 min readThreat reportPrompt injectionAgents

    Prompt injection in the wild: 21 incidents against deployed copilots and agents, 2025 to 2026

    A sourced record of real prompt-injection attacks on Microsoft 365 Copilot, ChatGPT, Gemini, Salesforce Agentforce, GitHub Copilot, Cursor, Perplexity Comet and others: how each worked, what leaked, which got a CVE, and what the pattern says about where the defence has to sit.

    Read the report

Get started

Ship the AI.
Keep the control.

The harness is up in under two minutes. Free to start, no card needed.

Harness listening · deploy in under 2 minutes

Start free Developer plan · 50K tokens/mo · 3 seats

Book a demo

We’ll walk it through with your stack in mind.

anupama@konsole.one, sheoli@konsole.one

A product of ClearTrust · Secure · Private · Model-agnostic