Skip to main content
Mallory
← Back to News
ai-platform-securitythird-party-vendor-breachinformation-technology-sector-threatleaked-secret-api-key

AI Canary Study Finds Evidence of Three Provider Breaches and Prompt Theft

Emerging Technology Risk · 7 sources · Updated 10h ago · First seen Nov 19, 2026

AI Canary Study Finds Evidence of Three Provider Breaches and Prompt Theft

Dragos researcher HD Moore reported that approximately two million AI prompts seeded with unique tracking URLs and canary credentials uncovered evidence of compromises at three unnamed AI hosting providers, unauthorized access to customer prompts by marketplace resellers, and apparent violations of privacy settings. The study, conducted between July and November 2026, tracked attackers using planted AWS credentials in tens of thousands of API calls to inspect cloud resources, attempt to enable additional services, and establish persistent access. The largest reported compromise captured requests during several short windows, including requests rejected before model execution. Resellers also allegedly accessed prompts through built-in logging. Dragos

The findings highlight risks from opaque AI routing chains: prompts submitted through U.S. providers reportedly reached an unnamed Chinese AI lab, which fetched secret URLs even from requests marked Zero Data Retention or “no data collection.” These observations raise concerns about confidentiality and data-residency assurances, although canary activations do not necessarily establish which intermediary leaked a prompt, and the unnamed providers limit independent verification. Dragos said affected providers were notified. Recommended mitigations include filtering secrets from prompts, minimizing intermediaries, avoiding reseller marketplaces, explicitly selecting upstream providers, and monitoring seeded canaries; on-premises inference is another option where feasible. Ars Technica

New surface, new adversariesTrack how attackers are adapting to this technology.
Track this risk
Event timeline

How this story unfolded

Events from the most recent confirmed update back to the earliest known activity.

20 events · Jul 2026 – Nov 19, 2026
  1. Reported Nov 19, 20262d ago

    Router says it stopped forwarding requests to the marketplace

    After notification, a router said it had stopped forwarding requests to the inference marketplace implicated in the research.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  2. Reported Nov 19, 20262d ago

    One affected provider publishes a preliminary security notice

    One breached provider published a preliminary security notice following the investigation. The references do not identify the provider or give a publication date for its notice.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  3. Reported Nov 19, 20262d ago

    Researchers notify providers of confirmed intrusions

    Researchers notified affected providers about the confirmed late-October intrusions. None of the three compromised hosting providers reportedly knew about the breaches before notification.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  4. Reported Nov 19, 20262d ago

    Route tracing exposes undisclosed intermediaries and prompt modifications

    Researchers traced approximately 2,000 inference routes, finding several hundred with unidentified systems and a handful sending images outside the advertised processing path. Findings included an undisclosed OpenRouter hop between SurplusIntelligence and Anthropic, a seller adding roughly 1,000 hidden tokens per prompt, and another fetching images through a shared consumer subscription.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  5. Reported Nov 19, 20262d ago

    Secret prompt URL appears in a Feishu company chat

    A canary URL appeared in a Feishu company chat 24 days after its originating prompt and was opened by six devices in Beijing and Shanghai within two minutes. Dragos said the prompt had been submitted through a U.S.-based provider with Zero Data Retention enabled; Ransomnews noted inconsistent descriptions of the originating provider.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  6. Reported Nov 19, 20262d ago

    Another provider revisits secret URLs days after inference

    A second provider's scraper repeatedly accessed seeded URLs days after the originating requests. The behavior occurred across requests delivered through different AI routers.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  7. Reported Nov 19, 20262d ago

    U.S. provider continues crawling a secret URL for more than two months

    One large U.S. provider continued accessing a secret URL more than two months after it appeared in a single request to the provider's search endpoint.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  8. Reported Nov 19, 20262d ago

    Git client attempts to clone repositories referenced in AI prompts

    A Git client operating on another cloud attempted to clone private repositories whose addresses appeared in seeded prompts. The attempts occurred exclusively for prompts involving the unnamed Chinese AI lab's models.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  9. Reported Nov 19, 20262d ago

    French residential user tries SSH passwords and AWS keys from prompts

    A person using a Mac on a residential connection in France manually attempted root SSH logins with passwords planted in two prompts, then tested several seeded AWS keys. One key had been submitted through a separate router that forwarded traffic to the marketplace without disclosing that hop.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  10. Reported Nov 19, 20262d ago

    Residential user opens canary links after marketplace processing

    Approximately a day and a half after two prompts passed through one marketplace seller, someone on residential broadband opened their secret URLs and attempted username guessing. Equivalent prompts processed by another seller did not trigger access.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  11. Reported Nov 19, 20262d ago

    Attackers use exposed AWS keys for reconnaissance and persistence attempts

    Attackers generated tens of thousands of AWS API calls to validate keys, list S3 buckets, and search Secrets Manager and Systems Manager Parameter Store. They also attempted to enable Bedrock models, create Bedrock API keys, and add IAM users for persistent access.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  12. Reported Nov 19, 20262d ago

    Two additional provider breaches expose Zero Data Retention requests

    Researchers found that the other two compromised hosting providers also exposed requests marked Zero Data Retention. Credentials submitted separately to two providers were rapidly used together from one rented server, suggesting a common adversary.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  13. Nov 19, 20263d ago

    Dragos publishes AI supply-chain investigation findings

    On November 19, HD Moore published findings describing evidence of three compromised AI hosting providers, reseller-associated prompt exposure, and apparent violations of privacy settings. The affected providers, marketplace, and Chinese AI lab were not named, limiting independent verification.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  14. Nov 2026

    Researchers begin reporting data-sharing issues to providers

    Researchers began reporting data-sharing issues in early November. Approximately a dozen providers became involved in follow-up concerning the investigation's findings.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  15. Nov 1, 202621d ago

    Attacker starts checking stolen credentials with ARTEX

    On November 1, an observed adversary began using the LLM-driven penetration-testing framework ARTEX for a small number of credential-validity checks. Researchers identified artex-agent and artex-judge-r121 user-agent strings in the activity.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  16. Oct 22, 20261mo ago

    Attacker begins using an AWS key planted the previous day

    An AWS key submitted in an AI prompt on October 21 was first abused on October 22. The credential remained in use three weeks later.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  17. Oct 2026

    Largest observed provider breach captures requests at intake

    During several short windows in late October, attackers reportedly stole every research request submitted to one provider, exposing a couple hundred planted AWS keys. Rejected requests were also exposed, indicating interception before model execution.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  18. Jul 2026

    Chinese AI lab accesses secret links from privacy-flagged prompts

    Between July and October 2026, an unnamed Chinese AI lab's systems fetched secret URLs from more than half of prompts sent to its models, including roughly two-thirds of privacy-flagged requests. Requests submitted through U.S. hosting providers triggered the same scraper, indicating that their prompt contents reached the lab.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  19. Jul 2026

    First provider investigation uncovers compromised internal credentials tooling

    A breach investigation beginning in mid-July eventually uncovered evidence that an attacker had compromised an AI provider's internal credentials tool. Associated login attempts continued into late October, when the provider still had not publicly disclosed the incident.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
  20. Jul 2026

    Researchers begin seeding AI prompts with traceable secrets

    Between July and November 2026, HD Moore and Dragos researchers submitted approximately two million prompts containing unique URLs, credentials, and other canaries to AI services. The researchers tracked subsequent access to identify exposure across providers and intermediaries.

    AI Supply Chain Security: What Canary Tokens Found | Dragos - Dragos
Do you run this?Follow how adversaries are adapting to this technology, and where it touches your stack today.
Track This Risk
Mallory platform

See the full picture, correlated to your attack surface.

This page covers what’s public. Mallory adds the parts that aren’t: which of your assets are affected, which threat actors are using it right now, and what to deploy next.

Exposure mappingMap indicators from this story to your assets and identify affected systems in minutes.
Threat actor evidenceEvery observed campaign, victim, and pivot linked to actors named in this story.
Associated malwareMalware, exploits, and IOCs connected to the activity described here.
Detection signaturesYARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Scheduled alertsGet matching new stories delivered to your team as they break, not the next morning.
AI threadsAsk questions about this story and take action on the answers.