The ninth Tallinn Digital Summit opened with a folk tale about keeping one end of the rope. Speakers from Estonia’s prime minister to Stuart Russell then located that rope in key custody, consequential testing and agent accountability records.
Tallinn Digital Summit opens on who holds the rope in the agentic era Read More »
Microsoft says it has identified two North Korean groups using ARTEX since late October and that their AI-enabled operations remain human-directed. Reports from three security firms show why the tool alone cannot identify an attacker.
Johns Hopkins Bayview responded to a shadow AI upload with a user-side deletion and a request that the tool’s operator delete the data. No account reviewed shows the operator did, and HIPAA leaves the hospital to document what it did.
NIST, the NSA and CISA advised developers of highly capable AI agents on Oct. 16 to preserve model weights, checkpoints and rejected tool calls for investigation when an agent takes “significant uninstructed actions” that compromise its controls. The agencies’ 14-page
The owner of ransomware-recovery firm MonsterCloud is charged with telling clients he had proprietary decryption tools while secretly paying their attackers. The allegations raise a question for every ransomware victim: does your own file show what was paid?
MonsterCloud indictment exposes the ransom payment a victim’s own file may not record Read More »
The Wikimedia Foundation says agents it believes OpenAI operates made unapproved edits, probed its Etherpad and sent heavy traffic to its data services. Its account adds a victim’s records to the growing record of agent incidents.
Wikimedia publishes its own account of suspected OpenAI agent activity on its wikis Read More »
The Cyber Law Toolkit’s 2026 update adds three scenarios: an artificial intelligence cyber defender that spoofs the red cross emblem, peacetime duties to protect civilians from cyber harm, and a harassment and deepfake campaign analyzed as torture.
Cyber Law Toolkit tests AI deception, cyber torture and peacetime cyber duties Read More »
Berlin refused an extortion demand on Aug. 28, and a ransomware crew posted an auction afterward. If the files publish, U.S. counsel will find that Model Rule 4.4(b) covers only what was inadvertently sent, and that the rest turns on a choice-of-law provision that may not point at home.
Berlin refused the ransom before the auction opened Read More »
Two House committees asked DoorDash for seven categories of records by August 14, covering every Chinese-developed AI model it has evaluated or used since January 2025. The request is not a subpoena, but it reaches further than a model list, and for the first time in this investigation it names the functions expected at the briefing that follows.
DoorDash AI inquiry turns model selection into a governance test Read More »
On a preliminary record, the Ninth Circuit made the user, not Perplexity, the party accessing Amazon under the CFAA and California’s CDAFA. The records of that access can sit partly with an outside operator that no litigation hold reaches by default.
Ninth Circuit vacates Amazon injunction; AI agent logs emerge as eDiscovery issue Read More »
Read the recent AI evaluation disclosures involving four labs as one repeated failure and you get it wrong. A novel exploit, a misconfiguration and a leaky allowlist are different problems, but they share a weak layer, an unanswered liability question, and a preservation gap that lands on legal and information governance teams.
Recent AI evaluation incidents expose gaps in containment, configuration and evidence Read More »
The European Commission fined Google €890 million and required compliance within 60 days. The harder problem is evidentiary: gatekeepers face growing pressure to document how ranking decisions were made, not simply to change them.
Google’s €890 million DMA penalty package starts a 60-day compliance clock Read More »
