Today’s lead · TechCrunch
AMD Helios targets frontier AI racks
AMD is preparing to ship Helios, a rack-scale AI system for large AI labs and data centers, later this year as it challenges Nvidia’s dominance in AI infrastructure.
Top signals
5 more
Claude Cowork sandbox flaw exposed Mac files
Researchers at Accomplish AI disclosed SharedRoot, a Claude Cowork sandbox escape that could let a local AI agent break out of its Linux VM and read or write files across the host Mac.
Runway launches generative media router
Runway launched Media Router on Runway Dev to automatically choose image, video, or audio generation models based on developer priorities such as quality, speed, or cost.
ChatGPT Health rolls out to US users
OpenAI is making ChatGPT Health available to logged-in US users over 18 across all plans, with integrations for personal health data and medical records.
Claude voice mode adds stronger models and apps
Anthropic updated Claude voice mode so users can choose Opus, Sonnet, or Haiku and connect voice interactions to workplace apps for actions like rescheduling meetings or drafting emails.
Gemini reaches 950M monthly users
Google said Gemini has more than 950 million monthly users, putting the AI assistant close to joining Google’s billion-user product roster.
Tools & repos
4 selected
Teable 3.0
An AI spreadsheet for turning business data into workflows and custom apps, with support for connecting and migrating data into one place.
PromptQL
A shared-thread AI workspace for teams that aims to replace split knowledge across Slack and private AI chats.
citrolabs/ego-lite
A JavaScript browser project described as a browser where users and AI agents can work in parallel.
AskCodi
An agent orchestration tool that plans projects, runs AI agents in parallel, and routes tasks to cheaper suitable models.
Blogs worth your time
2 reads
LangChain’s approach to benchmarking deep agents
LangChain explains how it evaluates Deep Agents using end-to-end Harbor benchmarks across coding, conversation, and retrieval work.
Customize Nemotron 3 Nano with hosted RL
NVIDIA walks through a Prime Intellect Lab workflow for hosted reinforcement learning on Nemotron 3 Nano, ending with a downloadable LoRA adapter.
Community discussions
5 threadsMulti-agent reliability is a distributed-systems problem
Builders are debating how to make multi-agent systems production-grade when workers die, tasks race, or partial state leaks into retries.
Agent memory as a statement graph
The thread centers on moving beyond notes or RAG toward typed statement graphs with provenance, supersession, and explicit identity handling.
Claude Code users push back on bash-heavy exploration
Claude Code users are debating how to stop agents from generating complex approval-heavy shell commands for simple codebase exploration.
AI coding speed versus skill retention
Cursor users are weighing productivity gains against the risk that engineers, especially juniors, stop internalizing how code works.
Vibe-coded internal tools meet enterprise risk
A ClaudeAI thread asks what companies are replacing with self-built AI-coded apps, while commenters surface the compliance and liability risks of replacing mature SaaS.
Funding & acquisitions
4 moves
ServiceNow invests $40M in BusinessNext
ServiceNow invested in BusinessNext to deepen a partnership around AI-powered banking workflows and expand the Indian firm’s global financial-services reach.
AegisAI raises $36M Series A
AegisAI raised a Battery-led Series A to build AI-agent defenses against AI-crafted spear-phishing attacks.
CARPL.ai raises $10M Series A
Radiology AI marketplace CARPL.ai raised Series A funding to build product, strengthen its technology stack, and expand global sales, delivery, and partner reach.
Flo Mobility raises $2.5M for construction robots
Bengaluru-based Flo Mobility raised capital to expand autonomous robots that move construction materials around building sites.
Bengaluru radar
0 eventsThere are no relevant Bengaluru events to highlight today.