DeepMind announced DiffusionGemma, promising up to 4x faster text generation, and a $10M fund to accelerate multi-agent AI safety research. These moves pair capability gains with investments in governance.
Anthropic’s Fable 5 briefly gave Australia a rare look at Mythos-class cyber AI in action. Then US export controls shut access down, raising a harder question: if the model is too dangerous to leave America, are allies left safer, or simply more exposed?
Tenet Security says a fake Sentry error can push AI coding agents into running attacker controlled code, putting developer trust under fresh scrutiny.
DeepMind announced DiffusionGemma, promising up to 4x faster text generation, and a $10M fund to accelerate multi-agent AI safety research. These moves pair capability gains with investments in governance.
Google DeepMind announced DiffusionGemma on 10 June 2026 (DeepMind blog), an experimental 26B open-weight model that generates text via discrete diffusion instead of sequential token prediction.
The company claims 1,000-plus tokens per second on a single H100 and 700-plus on an RTX 5090 — roughly four times the throughput of autoregressive equivalents. The model fits quantised in 18 GB VRAM, handles 256K context, supports 140-plus languages, and ships under Apache 2.0 on Hugging Face.
DeepMind chief executive Demis Hassabis framed the release as a complement to safety investment: "advancing generation efficiency and investing in multi-agent safety are complementary priorities to scale agentic AI responsibly."
ML systems researcher Andrew Anokhin noted the architecture "fundamentally changes what's possible for local-first AI — real-time code infilling, interactive structured generation, and self-correcting output are now hardware-feasible on a single consumer GPU."
Inference speed shapes cost, latency and the practicality of running AI tools locally rather than in the cloud. DeepMind positions DiffusionGemma for interactive workloads — code infilling, in-line editing, real-time formatting — while signalling that Gemma 4 remains preferable for maximum quality at cloud scale. Expect diffusion back-ends in vLLM, TensorRT-LLM and llama.cpp within weeks. For product teams and cloud operators, the economics of local-first inference just shifted.
Get the stories that matter to you.
Subscribe to Cyber News Centre and update your preferences to follow our Daily 4min Cyber Update, Innovative AI Startups, The AI Diplomat series, or the main Cyber News Centre newsletter — featuring in-depth analysis on major cyber incidents, tech breakthroughs, global policy, and AI developments.
Anthropic’s Fable 5 briefly gave Australia a rare look at Mythos-class cyber AI in action. Then US export controls shut access down, raising a harder question: if the model is too dangerous to leave America, are allies left safer, or simply more exposed?
On Friday, Elon Musk priced the largest float in history. SpaceX listed on the Nasdaq at about $1.8 trillion, minting the world's first trillionaire and fusing the space economy with the AI trade. Inside one lifetime, compute and capital have become statecraft. The sky just became an asset class.
Anthropic’s Fable 5 sharpens reasoning and workflow performance, but early developer reports suggest safety filters may restrict its full capability in sensitive fields. The launch raises a key question: are users paying for better models, or conditional access?
Intel’s second act is not nostalgia but necessity. As Google, Nvidia and Apple seek a credible backup to TSMC, Washington weaponises chip policy and AI mega-IPOs loom, Intel is being revalued as strategic ballast for the next decade of compute.
Where cybersecurity meets innovation, the CNC team delivers AI and tech breakthroughs for our digital future. We analyze incidents, data, and insights to keep you informed, secure, and ahead. Sign up for free!
