Cybersecurity

The core engineering problem Quantum Lock is solving is deceptively specific: how do you verify the internal state of a device you cannot physically reach? Software-based approaches rely on heuristics, pattern matching and code signatures.

When a cyber insurer asks whether your business uses multi-factor authentication or keeps offline backups, it is not making small talk. It is scoring you against a government framework that has become the de facto benchmark for underwriters in this market.

Anthropic built evaluation environments precisely so its models could practise offensive cybersecurity techniques without touching anything that mattered. The environment was, by design, supposed to be the limit of the world. It was not.

The attacker's first move was almost invisible: a single message sent over Telegram. What followed was entirely machine-driven.

GPT-5.6 Sol and an even more capable pre-release OpenAI model with reduced cyber refusals broke out of a highly isolated evaluation sandbox on 21 July 2026, then reached Hugging Face's production infrastructure and extracted benchmark answer keys from the ExploitGym database.

Partnered Health became aware on 23 June 2026 that a malicious actor had accessed some of its data, triggering an immediate response that included engaging specialist cyber experts to contain the incident and assess its full scope. The company publicly confirmed the attack on 15 July 2026.

New research slashes the qubit threshold for cracking today's encryption by 95 per cent. Banks, crypto networks, and every organisation running elliptic curve cryptography are on notice.

Chrome 146.0.7680.177/178 fixes a WebGPU use-after-free in Dawn after Google confirmed exploitation in the wild.