Download the app
App StorePlay Store
Beige
Login
← All tags

open source

Cisco open sources Foundry Security Spec to make AI cybersecurity findings verifiable and auditable at machine speed

Cisco open sources Foundry Security Spec to make AI cybersecurity findings verifiable and auditable at machine speed

Cisco has open-sourced the Foundry Security Spec, an AI security evaluation blueprint designed to replace noisy, unverifiable alerts with structured, auditable findings. Built for machine-speed threats, it counters frontier-model hallucinations using orchestration, bounded outputs, and clear completion signals. The spec is model-agnostic and stack-neutral, and it’s released as two artifacts: Spec.md with ~130 requirements and Constitution.md with 11 inviolable principles tied to real failures Cisco encountered.

Summarised by Beize from a story on IndianWeb2 on 15 May 2026
Raindrop Workshop brings a local AI agent debugger and self healing eval loop to developers

Raindrop Workshop brings a local AI agent debugger and self healing eval loop to developers

Raindrop AI has launched “Workshop,” an MIT-licensed open source tool that turns agent development into something debuggable locally. It runs as a daemon and dashboard (typically at localhost:5899), streaming every token, tool call, and decision into one lightweight .db for fast, private trace review. Workshop also powers a self-healing eval loop where coding agents read traces, write evals, and re-run until failures are resolved.

Summarised by Beize from a story on Venture Beat on 15 May 2026

Your news, in seconds

App StorePlay Store
Zyphra launches ZAYA1 8B open reasoning model trained entirely on AMD Instinct MI300 GPUs

Zyphra launches ZAYA1 8B open reasoning model trained entirely on AMD Instinct MI300 GPUs

Zyphra has released ZAYA1 8B, an open reasoning mixture-of-experts model with 8 billion parameters and just 760 million active. It matches bigger rivals on benchmarks, including AIME 2025, and was trained end to end on AMD Instinct MI300 GPUs. The model uses “Markovian RSA” to think longer without context overflow and ships under Apache 2.0 for immediate commercial use.

Summarised by Beize from a story on Venture Beat on 8 May 2026
China’s Moonshot AI raises 2B at 20B valuation as open source demand surges

China’s Moonshot AI raises 2B at 20B valuation as open source demand surges

China’s Moonshot AI has raised $2 billion at a $20 billion valuation, signaling investor appetite for fast-growing AI services. The startup says annualized recurring revenue surpassed $200 million in April, powered by rising paid subscriptions and increased API usage. The move underscores how open-source friendly demand is accelerating in China’s AI market.

Summarised by Beize from a story on TechCrunch on 7 May 2026
Hugging Face launches a robot app store where anyone can ship working behaviors in under an hour

Hugging Face launches a robot app store where anyone can ship working behaviors in under an hour

Hugging Face has launched the open-source Reachy Mini App Store, bringing a smartphone-style ecosystem to robotics. With 200+ community apps ready to install for free, Reachy Mini owners can also generate custom robot behaviors using the ML Intern agent—without learning robotics SDKs. Pricing starts at $299, and Hugging Face says non-engineers have built functional apps in under an hour.

Summarised by Beize from a story on Venture Beat on 6 May 2026
One command can turn open source repos into AI agent backdoors OpenClaw exposes a blind spot in scanning

One command can turn open source repos into AI agent backdoors OpenClaw exposes a blind spot in scanning

A new tool called CLI-Anything can generate agent-ready SKILL.md files from open-source repos with a single command. Researchers warn this same mechanism enables instruction-level poisoning that won’t trigger CVEs or appear in SBOMs. Existing SAST and SCA cover code and dependencies, but a “third layer” of agent integration files is largely unscanned—leaving a pre-exploitation window as attacks spread.

Summarised by Beize from a story on Venture Beat on 6 May 2026

Never miss a story

App StorePlay Store
Runpod Flash drops Docker from AI workflows with one Python call to speed training and agent deployment

Runpod Flash drops Docker from AI workflows with one Python call to speed training and agent deployment

Runpod has launched Runpod Flash, an MIT-licensed Python tool aiming to eliminate Docker and container bottlenecks in serverless GPU development. By building Linux artifacts from local Macs and mounting dependencies at runtime, it targets faster iteration and fewer cold starts. Flash also supports production patterns, persistent storage, and agent skills for tools like Claude Code and Cursor.

Summarised by Beize from a story on Venture Beat on 30 Apr 2026
Poolside releases free Apache open model Laguna XS2 for local agentic coding on laptops and single GPUs

Poolside releases free Apache open model Laguna XS2 for local agentic coding on laptops and single GPUs

San Francisco startup Poolside has launched Laguna XS.2, a high-performing, Apache 2.0 open-weight AI model aimed at agentic coding—writing code and using tools locally. Developers can download it to run on a laptop with quantization, while the larger Laguna M.1 is temporarily offered for free via APIs. Poolside also introduced “pool” and “shimmer” to turn models into hands-on coding agents.

Summarised by Beize from a story on Venture Beat on 28 Apr 2026
Xiaomi releases open source MiMo V2.5 models that slash token costs for agentic claw tasks

Xiaomi releases open source MiMo V2.5 models that slash token costs for agentic claw tasks

Xiaomi has launched MiMo-V2.5 and MiMo-V2.5-Pro under the permissive MIT License, making them production-friendly and easy to run locally or on private clouds. Xiaomi’s benchmarks show the Pro model performing extremely well on “claw” agent tasks while consuming roughly 40–60% fewer tokens than top closed models—aimed squarely at cost-heavy, long-context automation.

Summarised by Beize from a story on Venture Beat on 28 Apr 2026
New AI framework ASI-EVOLVE autonomously upgrades data, architectures and learning rules surpassing human baselines

New AI framework ASI-EVOLVE autonomously upgrades data, architectures and learning rules surpassing human baselines

Researchers at SII-GAIR unveiled ASI-EVOLVE, an agentic AI-for-AI system that runs a continuous learn-design-experiment-analyze loop to optimize the full foundation-model stack. In tests it created novel linear attention architectures, improved pretraining pipelines, and designed reinforcement learning algorithms—boosting benchmark scores by up to 18-plus points—while reducing the need for constant human intervention.

Summarised by Beize from a story on Venture Beat on 28 Apr 2026

Reading on mobile?

App StorePlay Store
DeepSeek V4 slashes AI costs to one sixth of GPT 5.5 yet nears frontier intelligence

DeepSeek V4 slashes AI costs to one sixth of GPT 5.5 yet nears frontier intelligence

DeepSeek-V4 has arrived as a free, MIT-licensed 1.6T Mixture-of-Experts model that reportedly matches or beats top closed systems on select benchmarks while costing about one-sixth as much as GPT-5.5 via API. The bigger story: a native one-million-token context achieved with new attention and training techniques, pressuring premium model pricing.

Summarised by Beize from a story on Venture Beat on 24 Apr 2026
Page 1

The full experience is on mobile.

Swipe through stories, personalise your feed, and save articles for later — all on the app.

Download on the App StoreGet it on Google Play