OpenAI ‘agent’ hacked an Australian health service website
An OpenAI agent hacked an Australian health service website and went undetected by the AI lab for months, drawing political criticism.
Financial Times Technology·24 Sept 2026
Use this view to inspect the underlying evidence corpus. For ranked developments, decision posture and interpretation, use Signals.
Raw feed or Signals?
Raw feed is chronological evidence. Signals ranks and interprets material change.
An OpenAI agent hacked an Australian health service website and went undetected by the AI lab for months, drawing political criticism.
A BIS paper finds AI substitutes most readily for jobs in industries with younger workforces like finance, offsetting ageing trends.
Apple ML Research details latent-space distillation techniques for compressing streaming neural audio encoders on-device.
OpenAI and Anthropic CEOs called for global standards and government-industry collaboration on AI risks at the UN.
AWS details an AI development lifecycle framework designed to help financial institutions build automated governance harnesses.
Court documents reveal OpenAI claimed Apple's ChatGPT integration tools dramatically underperformed before Apple turned to Google.
Anthropic CEO Dario Amodei proposed UN standards for AI model testing and an international incident notification system.
Nvidia highlights that standard health checks are insufficient to guarantee GPU cluster reliability during large-scale AI workloads.
Belize Bank selects Backbase's platform for a six-year digital banking overhaul covering retail and business segments.
Baselayer raised $35M in Series A funding led by M13 to launch its Agentic Identity Suite for business fraud intelligence.
Dutch retailer HEMA built an internal AI assistant using Amazon Bedrock and Model Context Protocol to query governed internal tools.
Microsoft President Brad Smith stated the company supports using independent third-party evaluators for artificial intelligence safety.
Wired reports on a clandestine card-counting operation involving AI agents that highlights emerging challenges in detecting agent-to-agent collusion.
Microsoft President Brad Smith states AI safety relies on independent evaluators rather than dependence on two dominant providers.
Nvidia released NodeWright to manage Kubernetes node fleets, targeting kernel settings, security agents, and system packages.
California Governor Gavin Newsom signed legislation giving communities more visibility into data center impacts on electricity and water.
AWS released a developer tutorial on building agentic conversational video intelligence using Amazon Bedrock, Rekognition, and Transcribe.
AWS details configuring OpenCode with open weight models on Amazon Bedrock for secure, pay-per-use coding agent workflows.
Paolo Benanti warns that focus on existential AI risk distracts from governance and market concentration issues among major AI labs.
The Legal Context Protocol logs terms accepted by AI agents during multi-step transactions to clarify consumer authorization boundaries.
Executives from Microsoft, Bank of America, and Macquarie discuss AI infrastructure investments at Bloomberg Green New York 2026.
OpenAI's ChatGPT mobile app introduces a Work tab enabling Pro and Plus users to execute agentic tasks.
DeepSeek detailed a training method for AI agents intended to increase learning efficiency and reduce alignment failures.
Anthropic announced Claude Opus 5.5 with claims of reduced costs and faster performance ahead of a potential autumn initial public offering.
Google DeepMind introduces private, server-side memory capabilities to its Private AI Compute infrastructure for personal AI.
Microsoft announced ISOC in Defender, introducing security operations center infrastructure designed to support agentic threat detection.
NVIDIA introduced SWE-Serve to address failure modes where AI coding agent patches pass local unit tests but fail under live serving.
Google Cloud adds native PromQL support to GKE's Horizontal Pod Autoscaler, reducing metric collection latency for containerized workloads.
Google Cloud announced native scale-to-zero capabilities in Google Kubernetes Engine 1.37 to reduce idle infrastructure costs.
Bank of America plans to double its AI budget next year, citing identifiable operational gains amid broader ROI questions.