Editorial section — [ 0x1 / sys ]
Software architecture, distributed systems, artificial intelligence and security.
OpenAI has just classified GPT-6 Astra at the 'Critical' level in cybersecurity. Behind this spectacular term lies a much more concrete shift: AI is beginning to transition from security assistant to genuine autonomous vulnerability researcher.
Fraudsters obtained the records of hundreds of Revolut customers without penetrating its servers. The incident reveals a deeper flaw: what happens when the channel meant to authenticate the state itself can no longer be considered trustworthy?
Chrome, Cloudflare, and major TLS stacks already protect part of your connections against future quantum computers. But key exchange, encryption, certificates, and the path to the origin server are not all at the same stage.
The best models promise immense productivity gains, but their APIs, logs, and tools create new trust boundaries. In 2026, the best trade-off between power and confidentiality is neither all-cloud nor all-local.
SpaceX wants to adapt NVIDIA's Vera Rubin systems to compute satellites consuming hundreds of kilowatts. Behind the dream of the space datacenter lies a much less spectacular problem: evacuating every watt of heat into the vacuum.
Firefox learns your history while Chrome can already navigate logged-in accounts, fill forms, and prepare purchases. This shift transforms the browser into an execution environment for AI agents and moves the boundaries of security with it.
With the Agents API, OpenAI is no longer just trying to produce better responses: it is building the infrastructure that allows an AI to retain its work, use a computer, delegate to other agents, and resume a task long after the initial prompt.
The unexplained disappearance of XChat from the App Store draws attention, but its protocol raises a far more interesting question: what does a "private" messaging app actually protect when metadata, key backups, and Grok enter the equation?