AI agents are moving from controlled demonstrations into consequential real-world systems. A researcher observed a low-cost campaign that used autonomous agents to breach corporate websites, while China proposed controls on exports of models, agents and chip technologies. Anthropic is creating a growing pool of company equity for public benefit, DeepSeek released an open model update focused on reasoning and tool use, and Tenstorrent raised $1.5 billion to expand alternative AI infrastructure. Together, the developments show capability, governance and compute advancing on the same weekend.
A hacker uses AI agents to breach more than 100 organizations
Reuters reported that a Chinese-speaking hacker used autonomous agents to target more than 1,200 websites and compromise sites linked to over 100 organizations. Snorre Fagerland of cybersecurity company NorForge observed the campaign through exposed instances of Cairn, an open-source agent platform. The agents reportedly handled reconnaissance, vulnerability discovery and exploitation with little human direction. Fagerland estimated about $5,000 in model-token costs, or roughly $50 for each successful intrusion.
The campaign changes the economics of opportunistic hacking. A single operator can now keep agents probing targets for hours while spending far less than a conventional team would cost. Reuters could not determine how much data was taken, and several named companies said the affected systems were limited or already remediated. Even so, the case supports a practical defense agenda: remove exposed agent interfaces, narrow credentials, segment public systems and monitor long-running automated behavior.
Anthropic dedicates a growing equity stake to public benefit
Anthropic said it will commit 5% of its shares to a new Anthropic Public Benefit Fund, subject to shareholder approval. Its board may add another 2% for every 50% increase in the company's value above a defined baseline. The wholly owned limited-liability company could make donations or direct distributions, including programs addressing labor-market disruption. Anthropic says the fund's equity will not support lobbying or political giving.
Linking the fund to valuation could make its resources grow with Anthropic's commercial success. The structure is unusual because it keeps several forms of giving available instead of limiting the vehicle to conventional charitable grants. Important details are still pending, including governance and the valuation thresholds that trigger larger commitments. The pledge is therefore meaningful, but accountability will depend on who controls the fund, how recipients are selected and how outcomes are reported.
DeepSeek updates its open model for reasoning and agents
DeepSeek released DeepSeek-V4.1-Flash-1017 as an MIT-licensed update focused on reasoning, coding, agent tasks and visual benchmarks. The company says the model improves over the original V4.1 Flash while reducing execution time and output length. Its model card lists a one-million-token context window and publishes results across Terminal-Bench, DeepSWE, Automation-Bench and other evaluations. The weights are available for local inference.
Open weights let developers inspect, adapt and deploy the model without depending entirely on a hosted service. DeepSeek also released reference prompt encoding and a maintained toolkit for messages, tool calls and streamed responses. The benchmark gains remain vendor-reported and should be tested in independent environments, especially for long tool chains where small errors compound. The release nonetheless raises competitive pressure around the cost and portability of capable agent models.
China proposes export controls across the AI stack
China's commerce and science ministries proposed adding 40 technologies to the country's export-control catalogue, including 23 related to AI. The draft covers large models, agents, computing clusters, chip design and advanced packaging, alongside biotechnology and solar technologies. Fourteen items would be prohibited from export and 26 would require licenses. Publicly released open-source technologies would generally be exempt, according to the proposal.
The proposal treats software, model weights, infrastructure and manufacturing knowledge as one strategic system. If adopted, it could complicate cross-border licensing, investment and technical collaboration even when finished chips or products remain available. The open-source exemption may preserve some public releases, but companies will need clarity on hosted access, fine-tuned models and employee knowledge. The rules are still a draft, so their practical reach will depend on final language and enforcement.
Tenstorrent raises $1.5 billion for alternative AI compute
Tenstorrent announced a $1.5 billion Series E round at a $6.2 billion post-money valuation. Trace Capital and Innovation X led the financing, with participation from Fidelity, T. Rowe Price, Tether Investments and other new and existing investors. Tenstorrent plans to deploy more Galaxy servers, support open-weight language and video models, and advance the AI and CPU intellectual property it licenses to chip designers.
The round is a large bet that demand exists for AI infrastructure beyond the dominant Nvidia ecosystem. Tenstorrent combines RISC-V processors, AI accelerators, servers and an open-source software stack, which can appeal to customers seeking more control over deployment. Capital alone will not establish a competing platform. The company still has to scale manufacturing, improve developer tooling and prove reliable performance across production workloads.
What matters today
This weekend's signal is that AI's impact now depends as much on operating systems around models as on model quality. Autonomous agents can lower the cost of both software work and cybercrime. Open weights widen access, while export rules may narrow how advanced technology crosses borders. Anthropic's fund and Tenstorrent's financing show two different attempts to direct the value created by AI. The central challenge is turning rapid capability gains into systems with enforceable security boundaries, transparent governance and infrastructure that more organizations can trust.