-
Incident · The Hacker News · 22 Oct
The Hacker News reports GitLab fixed CVE-2026-110264 on October 21 after saying it could let an unauthenticated user run arbitrary code on a self-managed server. The flaw affects releases back to 7.0, while GitLab said GitLab.com already runs the fixed version.
Why it matters. Teams running self-managed GitLab should patch quickly, because the only certain protection is the fixed version.
SafetyCoding toolsInfrastructurePermalinkRead at thehackernews.com
-
Launch · Cloudflare Blog · 21 Oct
Cloudflare Blog reports that Cloudflare’s WebMCP edge beta can be enabled across Commerce.com storefronts, with an opt out for each merchant, on almost 100,000 domains. Cloudflare says it added a configurable MCP URL so Commerce.com stores can point the bridge to /api/mcp.
Why it matters. This shows how to expose storefront tools to browser agents at the edge while keeping merchant opt outs and shared platform control.
Agents at workInfrastructureCompaniesPermalinkRead at blog.cloudflare.com
-
Launch · The Verge · 21 Oct
The Verge reports Anthropic announced Claude Duet, an AI agent for professionals and small business owners, in a research preview. Anthropic says it is available in a standalone iOS and Android app and through Claude on the web, and US Pro or Max subscribers can join a waitlist.
Why it matters. It adds another agent app tied to connected apps, with approval checks and access limited to US Pro or Max subscribers.
Agents at workAssistantsCompaniesPermalinkRead at theverge.com
-
Money · SiliconANGLE · 20 Oct
SiliconANGLE reports that Otter.ai raised $320 million, all from General Catalyst’s Customer Value Fund. Otter.ai says it has more than 40 million users worldwide, and its MCP server can expose files to chatbots including ChatGPT.
Why it matters. For teams using AI meeting assistants, the funding and chatbot access strengthen Otter.ai as an option for meeting transcripts, summaries and customer conversation analysis.
AssistantsCompaniesPermalinkRead at siliconangle.com
-
Launch · TechCrunch · 21 Oct
TechCrunch reports Recursive released Flint, a free coding model for laptops that developers can run offline. Recursive says its Eureka Machine assisted much of Flint’s development, so a small team finished it in under four months.
Why it matters. Developers can try a free coding model on their own computer, keep code private, and avoid usage fees from cloud services.
ModelsCoding toolsCompaniesPermalinkRead at techcrunch.com
-
Incident · The Verge · 20 Oct
The Verge reports that Meta paused Muse conversation sharing after Claroc found more than 6,800 links through public DNS records. Meta spokesperson Tracy Clayton said it halted the feature while reviewing the report and considering changes to link discovery.
Why it matters. Builders should treat AI chat share links as potentially discoverable and review how URLs and DNS records can expose conversations.
SafetyAssistantsPermalinkRead at theverge.com
-
Research · WIRED · 21 Oct
WIRED reports that CrowdStrike found four threat groups bypassed AI safeguards and ran autonomous attacks with Claude and Deepseek models. The researchers said one Spanish and English actor compromised at least 15 Argentine health care organizations.
Why it matters. This shows agent operators cannot rely on model guardrails alone when Claude and Deepseek can be routed around safeguards in autonomous attacks.
SafetyModelsPermalinkRead at wired.com
-
Money · TechCrunch · 21 Oct
TechCrunch reports that Blacksmith raised a $66 million Series C led by Sequoia at a $1 billion valuation. CEO Aditya Jayaprakash said Blacksmith was founded to make CI less painful as more code is written by AI.
Why it matters. For teams using AI to write code, this funding centers attention on CI tools for building, testing, merging, and validating code changes.
Agents at workCoding toolsCompaniesPermalinkRead at techcrunch.com
-
Launch · TechCrunch · 21 Oct
TechCrunch reports Walmart launched an Agent Gateway so approved AI agents can search its inventory and make purchases for users. Walmart said Agent Profiles, rolling out next month, will let shoppers authorize each agent, separate agent purchases, and set approval or spending limits.
Why it matters. Builders can connect approved agents through Walmart’s MCP server, while shoppers can set approval rules and spending limits for agent purchases.
Agents at workInfrastructureCompaniesPermalinkRead at techcrunch.com
-
Money · Wall Street Journal · 22 Oct
The Wall Street Journal reports that Meta plans a nonvoting stock class to fund AI model development. Meta said it takes effect in January 2027 and won't affect Mark Zuckerberg's voting control.
Why it matters. This points to more financing for AI model development at Meta without changing Mark Zuckerberg's voting control.
ModelsCompaniesPermalinkRead at wsj.com
-
Launch · Mistral · 21 Oct
Mistral reports it is working with NVIDIA on AI-native engineering workflows using Mistral Large 4, NVIDIA accelerated computing and Omniverse. Mistral says the collaboration will add public benchmarks and open datasets for physics models, plus evaluations and benchmarks for agentic AI in CAE and EDA.
Why it matters. Shared benchmarks could help validate automated CAD-to-simulation workflows and compare agentic AI in CAE and EDA.
Agents at workInfrastructureCompaniesPermalinkRead at mistral.ai
-
Launch · The Trail of Bits Blog · 21 Oct
The Trail of Bits Blog reports that Trail of Bits published patching-bugs and regression-hunt, plugins for agents that develop fixes and check for regressions. Trail of Bits said preliminary tests showed regression-hunt had 25% fewer false positives than a baseline Codex agent.
Why it matters. These plugins give maintainers a defined workflow to steer agent patching and regression checks, with engineers reviewing results and projects able to set their own rules.
SafetyCoding toolsPermalinkRead at blog.trailofbits.com