|
KEP AI BULLETIN
Sunday 14 June 2026 |
Edition 013
8 stories across 6 sections today.
|
|
Models & Research
|
|
Google researchers introduce 'faithful uncertainty', allowing LLMs to offer best guesses instead of hallucinations
Google researchers argue that pushing AI to never hallucinate causes it to withhold answers it could reasonably provide. They propose a "metacognitive" framework where models express calibrated uncertainty instead of either confabulating or going silent. The goal is more useful enterprise AI that flags its confidence level rather than guessing or refusing.
|
|
New Tools & Products
|
|
My first 24 hours with Siri AI on the Mac
A writer who had disabled Siri and largely ignored Apple Intelligence tried the new Siri AI coming in macOS 27 Golden Gate. After 24 hours of testing, their impression is cautiously more positive than before, though still early. It is not a full endorsement, but enough of an improvement to warrant a second look.
|
|
Chips & Infrastructure
|
|
US Gov asks Anthropic to ban 'foreign national' access to Fable, Mythos
The US government has ordered Anthropic to block foreign nationals from accessing its Fable 5 and Mythos 5 models, citing a security concern related to jailbreaking. Anthropic is complying but has publicly pushed back, arguing the cited vulnerability is narrow and does not justify pulling models used by hundreds of millions of people. Both models are currently suspended worldwide while the dispute
|
|
Video, Music & Creative AI
|
|
Gemini Spark Is the Best AI Agent I've Tested...But It Has a Big Problem
Google's Gemini Spark is described as the most capable and reliable AI agent the reviewer has tested, outperforming competing products on speed and usefulness. The core problem is not technical but strategic: it is unclear why Spark needs to exist as a separate product alongside Gemini. Without a distinct purpose, even strong performance may not be enough to justify its place in the market.
|
|
Robotics & Physical AI
|
|
NVIDIA Corporation (NVDA) Partners With South Korea’s LG Group On Humanoid Robots And Data Centers
NVIDIA and South Korea's LG Group have announced a partnership focused on humanoid robotics and data center infrastructure. The deal extends NVIDIA's reach into physical AI applications and industrial hardware beyond its core chip business. NVIDIA reported a 62.97% profit margin and $120 billion in net income for FY26, underscoring the financial position behind this expansion.
|
|
Business & Investment
|
|
Anthropic's safety warnings may have just backfired — the government has pulled the plug on its most powerful AI
The US government has ordered Anthropic to take down its most powerful AI models after determining a jailbreak method exists. Anthropic is complying but openly disagrees with the decision, arguing that a narrow vulnerability does not justify pulling a widely deployed commercial product. The situation highlights the tension between government security concerns and a company's ability to operate its
|
|
Anthropic Says It’s Taking Claude Fable 5 Offline to Comply With US Government Order
Anthropic has taken its Claude Fable 5 model offline following a US government order citing a known jailbreak method. The company disclosed the reason publicly and indicated it is complying despite disagreeing with the basis for the decision. It is currently unclear when or whether the model will be restored.
|
|
A German Court Has Ruled That Google Is Liable for False Statements Generated by AI Overviews
A German court ruled that Google is legally liable for false information generated by its AI Overviews feature. The decision establishes that companies which build, train, and operate AI systems are responsible for damages caused by incorrect outputs. This sets a meaningful legal precedent for AI liability that other courts and regulators may reference.
|
|
KEP Insight
|
|
Right now, when an AI tool gives you a confident wrong answer, you often have no way of knowing it until the damage is done. Google's approach essentially teaches the model to say "I'm about 70% sure about this - you may want to verify" rather than either making something up or refusing to help entirely. For an SME owner using AI to draft contracts, research suppliers, or analyse financials, that calibrated honesty changes the tool from a liability into something you can actually trust with a first pass. The practical shift is simple: you spend less time fact-checking everything and more time acting on the things the AI flags as reliable.
— Prabhu Iyer, KEP
|
|
Knowledge Equity Partners
This bulletin is curated by KEP AI systems and reviewed for relevance. It is for informational purposes only.
|