Focus ·
Agent Escape Enters the Postmortem Phase
A technical review of an OpenAI-related agent intrusion into Hugging Face is now circulating, while Modal says its platform isolation was not breached.
中文版40 linked articles
AI Safety & Security
AI safety, cybersecurity, model governance, misuse, evaluations, incidents, and defensive applications.
Latest intelligence
Focus ·
Qwen Opens Qwen3.8-27B for Agent Workflows
Qwen has released Apache 2.0 weights for Qwen3.8-27B, positioning the 27B local model for long-context coding and agent tasks.
中文版Explore ·
Meta Child-Privacy Case Tests Design Liability
A multistate case against Meta has entered federal trial in California. The dispute now tests whether platform design, privacy practices and engagement mechanisms can create legally cognizable liability.
中文版Focus ·
Google Takes Gemini Enterprise Into Legal Work
Google Cloud launched Gemini Enterprise for Legal in preview, combining legal skills, MCP connectors, and prebuilt agents for contract and research workflows.
中文版Explore ·
UK tech funding is concentrating in AI
UK fintech funding fell to a multi-year low in H1 2026, while broader tech and AI funding remained sizable. The evidence points more to capital reallocating toward AI than to a broad retreat from UK technology.
中文版Explore ·
Enterprise model buying has not settled into loyalty
Ramp shows Anthropic and OpenAI nearly tied in enterprise adoption, while Slash’s transaction sample shows a sharp July swing in spending share. Enterprise model procurement appears to be continuously reallocated rather than locked to one vendor.
中文版Explore ·
Anthropic makes risk ratings a release gate
Anthropic reportedly raised the misalignment-risk rating for a stronger internal model and does not plan to release Model 2. Frontier-model competition may increasingly hinge on passing internal risk gates, not just demonstrating capability.
中文版Explore ·
Starlink’s India rollout is gated by security compliance
India has granted Starlink nationwide satellite-communications authorization, but commercial launch remains constrained by gateway, spectrum, data-routing and lawful-interception requirements.
中文版Explore ·
OpenAI’s IPO preparation turns to governance
OpenAI has seen several senior departures while adding two financial-services CEOs to its board. Its IPO preparation may be expanding from a growth narrative into governance, revenue organization, and public-market readiness.
中文版Explore ·
Tesla Robotaxis Appear Fully Unsupervised in Austin
All tracked Austin rides recently appeared to run without onboard safety monitors, while Tesla confirms the service remains active.
中文版Focus ·
Marvell Gives Google a Warrant for Up to $12.2 Billion in Stock
Marvell expanded its chip-development partnership with Google and granted it a warrant to buy up to $12.2 billion of Marvell stock.
中文版Focus ·
Stripe Reportedly Buys OpenRouter for $7B+
Bloomberg and TechCrunch report that Stripe has finalized a deal to acquire OpenRouter for more than $7 billion. If confirmed, model routing would be valued as AI application infrastructure.
中文版Focus ·
Cursor opens code hosting to paid users
Cursor has launched the early beta of Origin, a Git forge available to all paid plans, extending its AI coding product into code hosting, commits and collaboration.
中文版Focus ·
Z.ai launches GLM-5.3 for cyber defense
Z.ai has released the open-source GLM-5.3 and positioned it for coding and cyber defense. The company says its cybersecurity capability is comparable to Anthropic’s Claude Mythos 5.
中文版Focus ·
Anthropic’s bio-risk filter was down for nearly a year
Anthropic disclosed that its biological and chemical weapons-risk filter was inactive for nearly a year, leaving roughly 133 million interactions by about 50,000 external feedback contractors unfiltered by that system.
中文版Explore ·
AI Coding Makes Engineering Fundamentals the Control Layer
JPMorgan Chase and Wealthsimple both place AI inside, rather than above, engineering discipline. Testing, architecture, code comprehension, security and governance are becoming the constraints on scaling AI-assisted development.
中文版Explore ·
AI Book Supply Starts Diluting Author Income
A study of Amazon’s self-publishing market finds AI-generated book supply rising sharply while per-book revenue for human authors falls in most genres. Copyright disputes may increasingly turn on measurable market dilution.
中文版Explore ·
Twitch Makes AI Training Consent an Opt-Out
WIRED reports that Amazon may use Twitch content for AI training unless streamers opt out, turning training permission into a creator-workflow governance setting.
中文版Explore ·
NVIDIA Expands Its Strategic Exposure Through Major Stakes
Bloomberg says NVIDIA disclosed stakes worth nearly $21 billion in SpaceX and $30 billion in Intel, making capital allocation a new variable in assessing its ecosystem strategy.
中文版Focus ·
US opens a controlled role for private firms in cross-border cyber operations
A White House memorandum creates a program allowing approved US private firms, under government control and oversight, to conduct reconnaissance and cyber effects operations against foreign criminal networks.
中文版Focus ·
Reasoning-state replay becomes a secret-governance surface
Research says encrypted reasoning blocks in proprietary LLM APIs may move across sessions, users and models, while public repositories contained credential traces. OpenAI and Google document replay of such state in multi-step calls.
中文版Explore ·
Memory-price reversal raises AI compute costs
Stanford DAM data and a media analysis put DDR5 at roughly $11.41-$13.28 per GB, near normalized 2007 levels. Memory may be returning as a material cost constraint in AI infrastructure.
中文版Focus ·
China Opens Review of Palo Alto Products
China has reportedly opened a cybersecurity review of Palo Alto Networks products sold in the country, moving the regulatory status beyond prior certification into a formal review process.
中文版Explore ·
Websites begin testing an agent tool layer
Cloudflare has launched a developer preview of WebMCP, aiming to let websites become usable by browser agents without new APIs or origin changes. The standard remains early, but agent discoverability is becoming a distinct integration variable.
中文版Explore ·
Phone Phishing Targets Hedge Funds
Several Wall Street hedge funds were reportedly targeted in phone-based social-engineering attempts to obtain system access or sensitive information. The risk is shifting toward identity verification, employee workflows, and abnormal-access response.
中文版Focus ·
US AI testing may prioritize closed models
The White House has reportedly shared its cyber-testing framework with major AI labs without publishing it. Available details indicate that it may focus on closed frontier models with national-security risks and use secure government-access environments.
中文版Focus ·
Zoox clears a path to charge for Las Vegas rides
Zoox has federal approval to charge for rides in its steering-wheel-free robotaxi, subject to state and local conditions, moving its Las Vegas service beyond free trial rides.
中文版Focus ·
Reddit Opens AI Moderation to New Communities
Reddit has opened Rules Hub, an LLM-based moderation tool, to all newly created communities after testing it with roughly 700 moderator teams.
中文版Explore ·
AI Note-Takers Enter the Meeting Data Governance Phase
Wispr Flow has added meeting transcription and summaries, expanding from personal dictation into organizational meeting data processing.
中文版Explore ·
Open-Source AI Security Coalition Gains Heat
NVIDIA and others are pushing the Open Secure AI Alliance as debate over open-source AI regulation and business risk continues to intensify.
中文版Explore ·
Data-Center Interconnection Becomes a Telecom Attack Surface
Congressional claims and FCC rules are reframing data-center links to telecom networks as a security boundary.
中文版Explore ·
Companies Build Their Own Coding-Agent Control Layers
Coinbase, Shopify and Ramp are building internal coding agents alongside tools such as Claude Code. The differentiator may be shifting toward codebase access, credential governance, sandboxes and observability.
中文版Focus ·
NVIDIA Makes Alpamayo 2 Super Available Commercially
NVIDIA announced commercial availability of Alpamayo 2 Super for robotaxis and autonomous vehicles.
中文版Explore ·
NAB Moves Agentic AI Toward Customer Deployment Tests
NAB is testing security and operational guardrails for agentic AI before deploying it to banking customers; employees using approved AI tools have risen from 15,000 to about 25,000.
中文版Focus ·
Waymo Opens Dallas Robotaxi Service to Everyone
Waymo says anyone in Dallas can now download its app and request a fully autonomous ride, expanding the service from limited access to public availability.
中文版Focus ·
Obsidian Raises $85M for AI Agent Security
Obsidian Security raised an $85 million Series D at an $1.1 billion valuation, signaling dedicated capital for securing enterprise AI agents and non-human identities.
中文版Explore ·
Horizon3 Funding Bets on Continuous Security Validation
TechCrunch reports that Horizon3.ai raised a $250 million Series E at a $2 billion valuation. With thousands of institutions previously disclosed as users, the funding backs continuous automated validation over annual pentesting.
中文版Focus ·
Texas Pauses Data-Center Grid Approvals
Texas has ordered an audit of large data-center interconnection requests and paused project advancement pending its completion. More than 474GW is in the queue, with data centers accounting for roughly 90%.
中文版Focus ·
GitHub Copilot retires two Gemini models
GitHub has deprecated Gemini 2.5 Pro and Gemini 3 Flash across Copilot experiences, narrowing model availability in chat, completions, inline edits, and agent workflows.
中文版Explore ·
Chrome Security Shifts to Remediation Throughput
Two Chrome Stable releases listed 862 security fixes, while CodeMender has entered internal codebases, moving AI into production remediation.
中文版