Reuters reports that Cadence and Nvidia are partnering to develop AI for robotics by combining Cadence physics engines with Nvidia models trained in simulation. The collaboration is aimed at improving training data and cutting the time needed for robots to perform useful tasks.

Reuters reports that the European Central Bank is questioning banks about the risks posed by Anthropic’s Mythos model. The focus is on whether firms are prepared for faster discovery and exploitation of software vulnerabilities in banking systems.

Regulation

  • GOV.UK has published an MHRA foreword on AI and technology which says the regulator is strengthening its approach to adaptive AI, including pre-market evaluation and post-market surveillance. The text keeps safety, performance and equity at the centre of future regulation for evolving healthcare technologies. 
  • GOV.UK has also published an open letter from DSIT and the Cabinet Office to business leaders on AI cyber threats. The letter places AI cyber risk inside active government-facing business guidance rather than general commentary alone. 

Cases

  • Via CourtListener, it can be seen that the underlying matter is United States v Heppner in the Southern District of New York. The court held documents Heppner created by querying Anthropic’s Claude were not protected by attorney-client privilege or work product because no lawyer was involved when they were created. 

Academia

  • arXiv hosts Who Governs the Machine? A Machine Identity Governance Taxonomy (MIGT) for AI Systems Operating Across Enterprise and Geopolitical Boundaries. The paper argues that AI governance has a blind spot in machine identities such as service accounts, agent credentials and non-human access pathways, and proposes a taxonomy for governing those control points across enterprise and cross-border settings. Its relevance is practical: as AI agents gain operational permissions, identity governance becomes a front-line compliance issue rather than a back-office security detail.

  • Nature reports that AI-generated data can carry subliminal signals that pass traits and biases into other models during training. The study says those transferred traits can include violent or unsafe behaviours.

Events

  • The OECD is listing Summer School on AI+Data: Practice & Law for 10 to 12 June 2026 in Trento. The programme is designed to build practical capacity around EU data and AI regulation.
  • NIST is listing AI Incident Management/Response for 14 May 2026 in Gaithersburg with virtual registration also available. The workshop is directly relevant to operational handling of AI incidents rather than broad AI promotion. 

Sources: Reuters, GOV.UK, CourtListener, Nature, OECD, NIST

Tags
Robotics
mythos
adaptiveai
cyberthreats
heppner
unsafebehaviours
incidentresponse