AMD FP8 Training Optimizations Merged Into PyTorch TorchTitan and TorchAO
AMD and Meta/PyTorch engineers have upstreamed FP8 training optimizations into pytorch/AO and pytorch/TorchTitan, enabling AMD Instinct GPU support with competitive FP8 performance out of the box. On Llama3-8B, FP8 training delive…
Weights & Biases Reaches One Billion Runs on Its Platform
Weights & Biases has reached one billion runs on its platform, according to a blog post published by CoreWeave. The milestone reflects cumulative usage of the MLOps platform across AI model and agent development workflows…
AI Agents Cross the Cyber-Offense Threshold — and Become Targets Themselves
The same agentic AI architecture that enterprises are racing to deploy for productivity is now operationally proven as both an autonomous offensive weapon and a high-value attack surface — and the evidence arrived in a single wee…
Distillation Becomes the Battlefield — and the Business Model
Distillation — the technique of extracting capability from large models into smaller ones — is simultaneously being weaponized as a geopolitical accusation, exploited through API security flaws, and commercialized as an open-source cost-cut…
ALL BRIEFS
RANKED · 10 THIS DAYEnterprise AI's Real Bottleneck: Orchestration, Not Models
The enterprise AI conversation has pivoted. The question keeping CIOs up at night is no longer which foundation model to deploy but how to govern, meter, and control the multi-agent systems already running in production — and the…
Coding Agents Go Multi-Agent: AWS, Tencent Cloud, and AgentRadio Ship Crew Architectures
The coding-agent market is converging on a new default architecture: not a single AI assistant paired with a developer, but coordinated teams of specialized agents running autonomously against shared codebases. **Why it matters*…
API Flaw Exposed Hidden Reasoning Traces from OpenAI, Anthropic, and Google Models
A newly disclosed vulnerability in the reasoning APIs of OpenAI, Anthropic, and Google allowed researchers to fully decode the encrypted chain-of-thought traces that frontier reasoning models generate internally — exposing not only propriet…
GhostJacking: Indirect Prompt Injection Hijacks AI Agents via Trusted Security Logs
Security firm Tenet has disclosed a new class of attacks called GhostJacking that exploits indirect prompt injection to manipulate AI coding and operations agents through trusted observability and security platforms, including Cloudflare, D…
Malicious LiteLLM Releases Exposed Thousands of Organizations via Supply Chain Attack
Two malicious releases of LiteLLM, the widely used AI-model proxy library, were published to PyPI in March carrying credential-stealing code that could harvest cloud keys, SSH keys, Kubernetes tokens, database passwords, and other secrets f…
ShieldBreak Zero-Day Bypasses Microsoft Defender Patch, Grants SYSTEM Privileges
A security researcher known as Chaotic Eclipse released a proof-of-concept exploit called ShieldBreak on August 12, targeting Microsoft Defender and granting SYSTEM-level privileges on fully patched Windows 10, Windows 11, and Windows Serve…