Anthropic Releases Claude Fable 5 and Mythos 5
On June 9, Anthropic announced two new models, Claude Fable 5 and Claude Mythos 5. Fable 5 is an attempt to make a Mythos-class model publicly available while ensuring safety, and it becomes the most powerful model Anthropic has ever offered for general use.
It achieves state-of-the-art results on nearly every major benchmark spanning software engineering, knowledge work, visual understanding, and scientific research. The longer and more complex the task, the more pronounced the gap over other models becomes. In a code migration project at Stripe, the model reportedly completed in a single day work that previously took a team two months.
The full-spec Mythos 5, meanwhile, will be provided only to trusted organizations. According to WIRED, because Mythos 5 carries misuse risks in the cybersecurity domain, Fable 5 includes a safeguard that falls back to Claude Opus 4.8 on certain topics. This safeguard is triggered in less than 5% of sessions on average.
Mythos 5 is also expected to find use in advanced research applications such as drug design, and it will be rolled out through a joint venture with Blackstone, an Anthropic partner organization.
WWDC 2026: Siri AI Fully Revamped, iOS 27 Announced
At WWDC 2026, Apple announced a complete overhaul of Siri, long considered a weak spot. In his keynote, Craig Federighi revealed that Siri AI is deepening its collaboration with Google.
In iOS 27, Apple Intelligence has been significantly upgraded, enabling Siri to handle more natural conversations and carry out complex tasks. At the same time, it has also been reported that Apple's in-house AI model contains no Google Gemini components, underscoring the company's two-track approach: leveraging partner technology while keeping its core model independent.
Security researchers have also raised concerns now that Siri can perform security-sensitive operations such as changing passwords via AI. On Hacker News, a discussion asked, "Apple's AI can now change your password — what could possibly happen?"
Agents' Last Exam: An Agent Benchmark Built by 300+ Experts Across 55 Industries
Featured on Hugging Face Daily Papers, "Agents' Last Exam (ALE)" is a new benchmark that measures how far AI agents can go in replacing the work of human experts. More than 300 experts constructed over 1,500 practical tasks spanning 55 industries.
The benchmark has three defining features. First, every task derives from work that experts actually performed on real projects, mapped to the US occupational classification system (O*NET). Second, it takes an "unconstrained" design that gives general-purpose computer-use agents full GUI + CLI access and evaluates results rather than methods. Third, it comes with objective evaluation criteria.
As an effort to measure agents' true capabilities in real-world work scenarios, it is drawing attention as a clear departure from conventional benchmarks.
EU Orders Meta to Bring Rival AI Chatbots Back to WhatsApp
The European Commission has ordered Meta to restore free access to competing AI chatbots on WhatsApp. The move appears to be taken under the EU's Digital Markets Act (DMA), a sign that regulation of platform operators favoring their own AI services is starting to have real teeth.
On Hacker News, a lively debate played out over how best to regulate big tech companies walling AI services off inside their own platforms.
Where Is the AI Jobs Crisis? — Debate Erupts
An analysis by investment firm Apollo titled "Where is the AI jobs crisis?" drew a large response, racking up 71 points and 84 comments on Hacker News. Amid widespread fears of AI-driven job displacement, the analysis argues that actual labor market statistics show no large-scale job losses.
Around the same time, a discussion titled "AI profitability is mathematically impossible" also attracted attention on Hacker News. The fundamental question — that massive investments in AI infrastructure have no visible path to monetization — has resurfaced once again.
TechCrunch also ran a piece titled "Can tech companies learn to love cheap AI models?", arguing that a shift toward more cost-efficient models could reshape the economics of AI.
Broadcom Announces $35 Billion AI Infrastructure Investment
Broadcom, together with Apollo and Blackstone, has launched a strategic AI platform on a 20-gigawatt scale and raised $35 billion in funding. With power demand from AI data centers surging, the move by a major semiconductor company to step directly into infrastructure building is drawing industry attention.
Separately, an Epoch AI analysis found that the AI boom has doubled computing infrastructure's share of US GDP — data suggesting that AI investment has already begun to have a non-negligible impact on the macroeconomy.