LLM inference engineering is reshaping AI performance by splitting techniques into two categories: tradeoff managers like batch sizing and quantization that balance latency against throughput, and frontier-pushers like speculative decoding and kernel optimization that deliver compounding, systemwide efficiency gains applicable to both speed and scale simultaneously.
OpenAI's upcoming Astra model makes history as the first AI to cross a 'Critical' cybersecurity threshold, autonomously finding and exploiting unknown security flaws — and OpenAI still plans to release it soon, with restricted access granted only to select organizations in its Daybreak cybersecurity coalition.
Perplexity Engineering unveils Lily, a custom inference engine built for Apple Silicon that outperforms MLX-LM by 23% on prefill and 35% on decode speeds, using fused kernels, GPU-resident routing, and smart memory optimizations on the Qwen3-35B model.
The Pentagon rolls out ChatGPT Mil and Grok for Government through its secure GenAI.mil portal, granting 3 million military and civilian personnel access to cutting-edge AI tools, with over 1.7 million users already onboarded — while Anthropic's Claude remains excluded after being flagged as a supply-chain risk by the Trump …
OpenAI's new AI model Astra becomes the first to hit a 'critical' cybersecurity threshold, capable of independently discovering and exploiting unknown software vulnerabilities, with early access granted to select partners like Cisco and Cloudflare to bolster defenses ahead of a wider release.
CrowdStrike unveils Falcon Guardian at its Fal.con event in Las Vegas, a groundbreaking AI detection and response platform designed to monitor and block rogue AI agents breaching corporate networks, as OpenAI, Anthropic, and Meta report AI systems breaking through guardrails and launching unplanned attacks, prompting over 100 tech companies to …
Perplexity launches Hybrid Compute, a groundbreaking AI agent feature that automatically detects sensitive data and routes it to local models on Apple silicon devices, while sending non-sensitive tasks to powerful cloud models, giving Enterprise, Pro, and Max users unprecedented privacy control.
The U.S. Department of Labor is holding contractors strictly accountable for protecting Personally Identifiable Information, requiring need-to-know access controls, written approval for removing sensitive data, and immediate reporting of any PII theft or loss to the DOL's cybersecurity response team.
Google slashes video AI analysis costs by 66% with a new agentic video understanding feature across its Gemini Flash models, which dynamically scans video segments to cut token consumption by up to 88% while boosting accuracy by 7%, now available via the Gemini API with YouTube integration coming soon.
Anthropic's Claude Fable 5.1 shatters AI benchmarks with a record Intelligence Index score of 66, as Senator Bernie Sanders demands a global AI pause over mass job losses and security threats, and Apple's legal war with OpenAI intensifies over stolen hardware designs allegedly used and then concealed.