[
    {
        "id": "INC-2026-008",
        "date": "2026-11-02",
        "name": "LLMjacking — Stolen API Keys, Million-Dollar Bills",
        "organization": "Google Threat Intelligence Group / Mandiant (victim enterprises across financial services, retail, technology, transportation)",
        "impact": "Up to $1,000,000 in Fraudulent Cloud Charges per Organization",
        "description": "Google's Threat Intelligence Group and Mandiant documented LLMjacking: attackers obtain enterprise LLM API keys through phishing, source-code leaks, misconfigured environments, and exposed client-side code, then consume the victim's paid compute at their own direction. Fraudulent charges exceeded $100,000 in a single day in individual incidents, with cumulative bills reaching $1 million per organization before access was revoked — spikes that arrive in hours rather than the slow drip of cryptojacking. Initial access vectors included leaked GitHub personal-access tokens, compromised service-account keys, MFA-bypassing session-hijacking portals, and CVE-2026-40217 in LiteLLM routing software. For any organization running inference behind metered cloud endpoints, the attack surface is a billing artifact that no key-rotation policy can close; sovereign local compute eliminates the prerequisite key entirely.",
        "source": "https://thehackernews.com/2026/11/google-warns-llmjacking-attacks-are.html"
    },
    {
        "id": "INC-2026-007",
        "date": "2026-02-17",
        "name": "Public Cloud AI Waives Attorney-Client Privilege",
        "organization": "United States v. Heppner (S.D.N.Y.)",
        "impact": "Client Work Product Turned Admissible Evidence",
        "description": "In United States v. Heppner, No. 25 Cr. 503 (JSR) (S.D.N.Y. Feb. 17, 2026), Judge Jed S. Rakoff held that neither attorney-client privilege nor work-product protection covers documents and communications generated through publicly hosted cloud AI platforms. The ruling established that submitting client information to third-party APIs like Claude or ChatGPT constitutes voluntary disclosure to a commercial vendor — instantly waiving privilege that took years of diligent counsel to build. For law firms, a single partner pasting facts into a public chatbot can make the entire matter's work product discoverable by prosecutors.",
        "source": "https://www.akingump.com/a/web/ssTGsd5NHbtZ1onzXQMTye/1_25-cr-503-27-memorandum.pdf"
    },
    {
        "id": "INC-2026-005",
        "date": "2026-05-29",
        "name": "The $500 Million Uncapped Seat Burn",
        "organization": "Unnamed Enterprise Client (via Axios)",
        "impact": "$500,000,000 Claude Bill in 30 Days",
        "description": "An AI consultant revealed to Axios that one of their enterprise clients accidentally racked up a staggering half-billion dollar bill in a single month on Claude AI after failing to configure usage and spending limits on corporate employee licenses. This represents the single largest cloud token billing disaster in internet history, showing why uncapped local compute is a financial necessity.",
        "source": "https://www.fastcompany.com/91550884/claude-ai-costs-climb-company-spent-half-a-billion-dollars-in-a-single-month-report"
    },
    {
        "id": "INC-2026-006",
        "date": "2026-09-29",
        "name": "The Subscription Treadmill: More Cost, Less Compute",
        "organization": "OpenAI",
        "impact": "Top Tier +67% Price with Halved Compute Allowance",
        "description": "At DevDay 2026, OpenAI introduced a $500/month Pro 500 tier (its prior top plan was $200/month, roughly $6,000/year per seat) and simultaneously reopened the $200 Pro plan with half the compute allowance, dropping the limit to 10x the Plus plan (from 20x) and cutting GPT-6 Pro chat messages from 200 to 100 per week. The episode crystallizes the public-cloud dependency trap: platforms control pricing, ration capacity at their discretion, and extract steadily more rent for shrinking allotments, compounding year over year while local sovereign compute costs stay fixed.",
        "source": "https://www.businessinsider.com/chatgpt-new-plan-pro-500-cost-compute-allowance-2026-9"
    },
    {
        "id": "INC-2024-001",
        "date": "2024-02-15",
        "name": "Legal Liability for Hallucinations",
        "organization": "Air Canada",
        "impact": "Court Order & Brand Damage",
        "description": "A passenger asked Air Canada's passenger service chatbot about bereavement policies. The chatbot hallucinated a custom policy promising a retro-active refund. The airline argued in court that it could not be held responsible for its chatbot's 'independent information.' The court ruled against the airline, establishing a legal precedent that companies are responsible for any hallucinations and promises made by their automated agents.",
        "source": "https://www.wired.com/story/air-canada-chatbot-refund-policy/"
    },
    {
        "id": "INC-2023-011",
        "date": "2023-07-18",
        "name": "System Performance Decay (Drift)",
        "organization": "Enterprise AI Users",
        "impact": "Silent Workflow Failure",
        "description": "A joint research study by Stanford and UC Berkeley verified that public model performances change significantly over time. For example, GPT-4's accuracy on prime-number identification tasks plummeted from 97.6% in March to just 2.4% in June. This 'silent drift' causes deterministic enterprise workflows built on public APIs to break without warning or error codes.",
        "source": "https://arxiv.org/abs/2307.09009"
    }
]