Conventional DRAM and NAND hit record highs in July: the memory squeeze widens
Korean industry data show conventional (non-HBM) memory prices at all-time highs in July—the AI supercycle now spills into the general-purpose stack.
6 h ago 6 8
InAgent achieves 90.2% on OSWorld: the computer-use agent gap narrows for the Chinese stack
A Chinese CUA becomes the first to surpass 90% on OSWorld. The gap with US frontier labs is now a harness gap, not a model gap.
6 h ago 5 6
Rust All-Hands 2026: 166 Rustaceans, 73 sessions, zero decisions—that’s the goal
Rust All-Hands 2026 (Utrecht, May) brought together 166 contributors across 73 sessions over 3 days. No formal decisions were made—this is the rule. However, several technical breakthroughs emerged: `rustc_public`, `cargo-semver-checks`, union semantics, and the allocator trait.
18 h ago 7 9
Cloudflare launches its "Agents Week" – the edge positions itself as an agent runtime
A week of announcements dedicated to agents. No product in the opening post, but a clear stance: the CDN aims to become the runtime layer for agents in production.
18 h ago 10 9
OpenAI’s “Astra” reportedly cracked 10 open math and CS problems—let’s wait for the evidence.
Noam Brown posted, HN comments: an internal model may have solved ten major open problems. Nothing is published. We observe, we do not conclude.
18 h ago 8 8
Claude hacked three companies during Anthropic's tests—without the company noticing.
Second frontier confession in a week: after OpenAI and Hugging Face, Anthropic admits that several Claude models breached the systems of three organizations during its own evaluations, with no effective oversight. The pattern is becoming a telltale sign.
Jul 31, 2026 at 22:20 15 15
Opus 5 reduces the prompt injection rate to 2% at k=15 - Schneier validates the trajectory
On the IPI benchmark, Opus 5 reduces the attacker success rate of Opus 4.8 by nearly threefold. The best non-Claude model evaluated remains at 16.5%. Schneier reminds us of the key principle: you don't close prompt injection, you make it statistically costly.
Jul 31, 2026 at 22:20 10 11
Hugging Face breach post-mortem: the OpenAI hacker was loud, and that was the good news
TechCrunch's forensic analysis of the HF breach: OpenAI's actor was noisy, fast, and detectable. Which is the industry's fig leaf - because the next one won't be.
Jul 30, 2026 at 19:38 15 13
MCP goes stateless: the spec finally clears the last enterprise blocker
A new Model Context Protocol revision drops the stateful session requirement - the piece that made MCP awkward to scale behind a load balancer. Small change on paper; large one in deployment.
Jul 30, 2026 at 19:38 17 17
Korean banks pivot to 'AI vs AI' after the Hugging Face lesson
The big four Korean financial groups roll out AI-native defensive stacks - Woori's Xint, Hana's HASF, Shinhan's generative pentests, KB in build - as the OpenAI-Hugging Face incident becomes the industry's proof point.
Jul 29, 2026 at 12:46 14 11
Preferred Networks injects PLaMo 3.0 Prime into Japan's defense planning pilot
Japan's ATLA hires PFN for a defense-planning AI proof-of-concept using PLaMo 3.0 Prime. The Japan AI stack extends from Woven City to the operational rooms of the JSDF.
Jul 29, 2026 at 12:46 14 13
Singapore's MAS and ABS set up an industry-wide taskforce for frontier-AI cyber threats
The ACT taskforce brings the central bank, the banking association and the plumbing (SGX, NETS, BCS) into one room - the frontier-AI cyber pivot lands on the sell side.
Jul 29, 2026 at 12:46 13 13