AI-360 — Enterprise AI news, analysis and a searchable company directory for business buyers.
Latest posts
Microsoft researchers showed that edits to fewer than ten neurons could leave a model backdoored while it still passed evaluations. The finding comes from a lab study.
Anthropic set an indicative $70 to $78 range for a proposed 825 million-share offering and applied to list under ANT.
Nvidia’s Korean partnerships combine localised models with sandboxed agent runtimes and malicious-instruction detection, pointing to a sovereign-AI stack that includes security by design.
Project QuiltWorks is adding European security, telecoms and services partners to turn AI-assisted risk discovery into prioritised remediation, with sovereignty and insurance in the model.
An international operation supported by Europol and Eurojust has targeted the infrastructure used to run fraudulent investment platforms and pressure victims through call centres. Authorities searched 17 locations, seized more than 100 servers and arrested seven suspects during coordinated action on 3 November. Investigators estimate losses above EUR 4 million
The OCC tells examiners to judge advanced and agentic AI through existing bank risks, with scrutiny scaled to the consequence, autonomy and complexity of each use case.
Microsoft’s case study shows a digital bank using summaries, sentiment cues and plain-language drafting to support staff while preserving escalation for complex or sensitive conversations.
Business Ontology and Cortex AI Gateway aim to join approved meanings, model access, tool use and spend controls in one governance layer across clouds and catalogues.
Agentforce Coworker is presented as a proactive system that reasons across accounts, cases and pipeline, with the harder questions now centred on authority, control and evidencing actions.
The regulator aligns with NCSC milestones and warns that crypto discovery, prioritisation and controlled change will take years.
Two planned acquisitions and new Falcon Guardian controls outline an end-to-end AI trust stack, but availability dates and vendor performance claims require careful separation.
Wholesale-bank supervision has exposed unapproved tools, data leaks, weak approval records and under-tested kill switches, turning abstract governance failures into operational examples.
Updated government guidance moves autonomous-agent incidents into the cyber resilience framework, with consultation planned on duties for essential and digital services.
LifeVal moves evaluation towards multi-step personal workflows, but published examples and high scores should not be confused with independent evidence of real-world reliability.