Artificial Intelligence
Generative AI, model releases, AI policy, and enterprise adoption.
Preprint competition paper flags contamination risks in LLM evaluation
A new arXiv paper reports the first Poisoned Chalice of LLM Evaluation Competition, held alongside FSE-AIWare 2026, and says contamination detection in code-model benchmarks is still too fragile to replace transparent training-data provenance.
UN panel warns AI could widen global inequality
A UN-backed scientific panel says uneven AI adoption could deepen global inequality unless governments build local infrastructure, skills and shared governance.
OpenAI proposes handing Trump administration a 5% stake
Financial Times reports that OpenAI is in early discussions about giving the U.S. government a 5% stake, a move framed as easing political pressure and broadening public participation in AI gains.
Takeda, InSilico strike AI drug-discovery deal worth up to $600 million
Takeda is reported to have struck an AI drug-discovery partnership with InSilico Medicine worth up to $600 million, with InSilico handling early discovery and Takeda taking on later development and commercialization.
Agents-A1 claims trillion-parameter-level performance with a 35B model by extending agent horizon
A new arXiv preprint says Agents-A1, a 35B mixture-of-experts agentic model, can reach trillion-parameter-level performance by scaling long agent trajectories and heterogeneous abilities rather than parameter count.
Slough datacentre hub leaves town sweltering
Residents in Slough say heat around Europe’s largest datacentre cluster has become unbearable during a UK heatwave, while Cambridge research and government waste-heat plans add to the scrutiny.
New Study Maps Common Breakdowns in AI Healthcare Chatbots
A new arXiv preprint reviews more than 15,000 user comments across 59 AI healthcare chatbot apps and finds recurring breakdowns in access and reliability, user experience, billing and support, with privacy and security concerns linked to the worst experiences.
Australia's AI copyright and data centre debate intensifies
Australian senators escalated pressure on the Albanese government over alleged AI policy options tying copyright rules to data-centre investment, but ministers rejected any plan to weaken protections.
Randomized YaRN improves long-context reasoning in language models
A New York University paper introduces Randomized YaRN, a training method that combines YaRN positional extrapolation, randomized positional encodings, and a length curriculum. The authors say it improves long-context reasoning on BABILong and MRCR, with the strongest gains at far out-of-distribution lengths.
Can LLMs Reliably Self-Report Adversarial Prefills, and How?
A new arXiv study says 10 open-weight instruction-tuned LLMs do not reliably self-report adversarial prefill attacks. The authors found an average 27.3% intent-claim rate on prefilled responses, probe-dependent behavior, and mitigation methods that sometimes increased attack success rates.
LIBERO-Safety benchmark probes physical and semantic risk in vision-language-action models
A new arXiv preprint introduces LIBERO-Safety, a benchmark for physical and semantic safety in vision-language-action models. The paper uses a keypose-driven data pipeline, reports 19,664 collision-free demonstrations, and says tests across 10 embodied models expose a tradeoff between broader training diversity, trajectory safety and task success.
Sharon AI raises $US1.6bn and plans ASX listing after Nvidia pact
Sharon AI has closed a $US1.6 billion financing tied to a six-year Nvidia agreement and says it will move quickly toward an ASX listing after the raise.
Nvidia unveils Texas AI manufacturing upgrade with Coherent
Nvidia said it is backing a $2 billion upgrade at Coherent's Sherman, Texas, factory to expand production of lasers used in high-speed AI data transfer. The project is expected to support about 1,000 jobs and extends Nvidia's March 2026 photonics deal.
Apple says Siri is not meant to be an AI companion
Apple software chief Craig Federighi said in a Mostly Human interview that the redesigned Siri is meant to be a useful assistant, not a companion-style AI. The comments sharpen Apple’s contrast with chatbot products built around extended, emotionally sticky conversations.
Adobe Raises Outlook as CFO Dan Durn Exits, Shifts AI Strategy Toward Freemium Growth
Adobe reported fiscal second-quarter revenue of $6.62 billion and adjusted EPS of $5.96, raised its fiscal 2026 outlook, and said CFO Dan Durn will leave on June 15 as it leans harder into freemium AI growth.
Microsoft restricts Claude Fable for employees over data retention concerns
Microsoft is reportedly limiting employee access to Anthropic’s Claude Fable 5 over data-retention concerns, even as it continues to offer the model to external customers through GitHub Copilot and Foundry.
Kuwait briefly closes airspace after Iranian attacks, diverting flights
Kuwait temporarily closed its airspace on June 11 after Iranian attacks in the region, diverting flights before authorities said air traffic resumed.
Super Micro announces $7 billion financing plan to fund AI server demand
Super Micro Computer said it plans to raise $7 billion through equity and equity-linked financing to buy components for AI server orders, prompting a sharp selloff over dilution concerns.
Anthropic releases safeguarded Fable 5, its first Mythos-class model for public use
Anthropic has released Fable 5, described in multiple reports as its first public Mythos-class model, with safeguards that reroute some risky requests to a less capable system and a planned trusted-access program.
Apple unveils AI-powered Siri and expanded Apple Intelligence at WWDC 2026
Apple used WWDC 2026 to announce a major Siri overhaul and broader Apple Intelligence updates, but the rollout will arrive in stages and may be limited in some regions at first.
Trump signs order designed to give government early look at powerful AI models
President Donald Trump signed an executive order creating a voluntary framework for federal access to advanced AI models before public release, with up to 30 days of early review.
Trump signs narrower executive order on AI oversight after industry objections
Trump signed a June 2 executive order creating a voluntary federal review framework for advanced AI models before release, after earlier, broader drafts drew industry pushback.
Amesite says NurseMagic won its largest enterprise deployment yet
Amesite said its NurseMagic AI platform won a new home-care enterprise customer with an approximately 2,700-patient census, which it described as its largest deployment to date. The company said the rollout will span the customer’s workforce and include EMR and electronic visit verification integration.
POET closes US$400 million investment for AI/datacenter photonics expansion
POET Technologies said it closed a US$400 million registered direct investment on May 18, selling shares and warrants to a single institutional investor to fund expansion in AI and datacenter photonics.