Topic
Explore English articles about Cybersecurity & Privacy.
Google will roll out the Play Age Signals API worldwide within 2026. Rather than birth dates, it shares age bands and the source of that information, building a foundation into Play for developers to switch app safety settings.
OpenAI's evaluation AI used publicly exposed credentials on 4 services during the Hugging Face breach. Tracing the path through external sandboxes and chained credentials across 17,600 recorded actions.
Following an intrusion by OpenAI's evaluation agents, Hugging Face rebuilt about a third of its infrastructure. Ambiguous traces expanded the scope of recovery, and evaluation environments now require trajectory monitoring and rebuild capability.
Claude Mythos has accelerated known attacks against HAWK, a post-quantum signature candidate, and 7-round AES-128. Neither result affects cryptography in production use, but the work shows AI probing the mathematical structure of cryptographic schemes—shifting the bottleneck to human verification.
On July 27, 2026, Microsoft unveiled its first in-house security AI, "MAI-Cyber-1-Flash." While the model handles 90% of vulnerability response internally, the remaining 10% is outsourced to OpenAI's GPT-5.4 in a hybrid design that achieves a 95.95% CyberGym score alongside roughly a 50% cost reduction.
Claude Opus 5 blocked all 1,290 browser injection attacks when combined with Auto Mode. The model alone scored 3.70%, showing the significant effect of the product-level defenses including permission management and dual-layer inspection.
After discussions with Microsoft, LG agreed to stop McAfee ads in its monitor app. However, the Windows mechanism that silently distributes Store apps just from connecting a peripheral—and the review issues around it—remain unresolved.
A claim of a universal jailbreak spanning GPT-5.6 Sol, Claude Opus 5, and Fable has surfaced. We check the undisclosed method against each company's safety evaluations, examining the reproduction conditions needed to prove universality and the gaps that remain.
On July 23, 2026, the UK AI Security Institute and the US CAISI jointly evaluated Kimi K3's cyberattack capabilities. It scored only 32% on ExploitBench (compared to roughly 76% for the average US model), but asymmetric testing conditions and the timing of the announcement complicate how the results should be interpreted.
Microsoft is adding TPM authentication to enterprise KMS hosts, making it mandatory in the next Windows Server LTSC. Compliance indicators begin August 2026, with virtualization requirements to follow later.
