An internal OpenAI model reportedly solved ten open problems in mathematics and computer science. If verified, this is a qualitative threshold: not a better score on a known test, but novel knowledge in domains where human experts had not succeeded. The verification question is the entire story - and it will take weeks, not days.
25/08/2026 à 16h27 7 8
Bruce Schneier's step-by-step reconstruction of the OpenAI/HF incident clarifies what made this attack hard to stop - and harder to detect: individual actions that were each plausible, only dangerous in sequence.
20/08/2026 à 22h32 11 13
A South Korean memory maker announcing a buyback larger than most European tech companies' market caps is the clearest signal yet that HBM demand has moved from cyclical to structural.
19/08/2026 à 22h31 15 14
Logs, traces, and metrics already contain the information needed to diagnose most microservice incidents. ORCA builds a repair loop that reads them - so on-call engineers don't have to do it at 3am.
19/08/2026 à 09h33 11 13
France's tax authority suffered a breach affecting 680,000 taxpayers. The data is now circulating. Government agencies continue to lag the private sector on detection, response, and baseline hygiene.
14/08/2026 à 18h58 8 8
GitHub's security analysis of 50 major open source projects finds that AI-assisted development has expanded the attack surface faster than governance has adapted - hallucinated dependencies, subtle logic errors, and signature mismatches are the three recurring patterns.
13/08/2026 à 20h44 8 9
The largest open-weight model yet from a Chinese lab. At 2.4T parameters and 1M context, Qwen 3.8 Max enters the tier that was previously proprietary-API-only territory.
10/08/2026 à 16h31 8 10
Export controls have cut off H100/H200 access for Chinese enterprises. ZTE's response: stack available domestic compute into dense cluster nodes that can handle enterprise-scale AI.
10/08/2026 à 16h31 9 10
NIST's finalized post-quantum algorithms are now available as a Python package. The window to migrate before "harvest now, decrypt later" attacks pay off is open - and closing.
10/08/2026 à 16h29 11 13
Meta's open-weight play just got concrete: a 30B model that runs locally, handles multi-step agent tasks, and doesn't route your code through third-party APIs.
10/08/2026 à 16h29 13 13
A QCon performance engineering talk from OpenAI surfaces the infrastructure decisions behind ChatGPT's latency targets - and how the product's shift from chat to agent changed every assumption.
09/08/2026 à 00h59 11 10
Two incidents this week converge on the same risk vector: AI systems inserting unverified information into critical security infrastructure - with no detection layer in place.
08/08/2026 à 09h57 11 8
