Firehose

China's Ministry of Commerce (MOFCOM) and Ministry of Science and Technology (MOST) have issued Announcement No. 48 of 2026, proposing the addition of 40 technologies to the Catalogue of Technologies Prohibited or Restricted from Export, including 23 in artificial intelligence. The announcement imposes an interim pre-export registration requirement for exports of restricted technologies, which applies now, including to exports that have started but are not yet complete. AI summary

AI-generated wikis are flooding the internet with inaccuracies, competing with volunteer-run wikis for search engine rankings and overwhelming them with traffic, potentially causing site instability and driving away contributors. This is exacerbating the challenges faced by independent game wikis, which already struggle to maintain profitability and attract and retain volunteers. The issue is part of a broader problem with generative AI's impact on the availability of information. AI summary

A cybersecurity startup, Cathedral, gained unauthorized access to Anthropic's internal systems in under 12 hours, exposing employee details, salary information, and internal messages. Cathedral used AI agents to find weaknesses in business software, and the incident was discovered during a security test. Anthropic confirmed that the researchers accessed a limited set of its sales, human resources, and IT systems. AI summary

Proton's CEO, Andy Yen, cancelled a company-wide "re-interviews" program that aimed to assess employees' use of AI, following an investigation by 404 Media that revealed security concerns and a culture of fear within the company. The re-interviews were intended to fire employees who did not use AI enough, and Yen acknowledged a "communication failure" but stated that the direction had not changed. The move is seen as damage control, and Proton users are questioning the company's commitment to security. AI summary

Large Language Models (LLMs) claim to collect and assemble a vast repository of creative works to generate unlimited, uncredited, and uncompensated imitations of user-created content. They argue that their vast computing power can churn out high-quality imitations, but acknowledge that "hallucinations" (incorrect or misleading information) are an inherent aspect of their technology. AI summary

Cisco's president has given a 27-year-old engineer, Padmanabhan Krishnamurthy, unlimited AI tokens to rewrite 98% of the company's legacy code within 24 months, with a goal of challenging the status quo and bringing fresh ideas to the organization. This approach aims to inject new ideas and challenge the existing corporate structure, but also carries significant risks, including potential technical issues and intellectual property rights concerns. The project's success depends on the engineer's ability to refactor the code effectively and test it thoroughly. AI summary

Sashank Gondala, CTO of FurtherAI, shares lessons from building an AI-native engineering org, focusing on optimizing the shipping process. Key changes include: - Agents writing code, tests, and reviews, with human reviewers focusing on judgment. - Parallelized testing and automated bug intake. - Rules-based checks for tribal knowledge. - Low-risk changes merging without human involvement. - Engineers taking ownership of planning, execution, and review. - Managers focusing on understanding product and customer needs, rather than just reviewing code. AI summary

Martin Fowler warns that Large Language Models (LLMs) can subtly alter writing styles, leading to an "uncanny valley" effect when readers encounter LLM-generated content. To avoid this, authors should not rely on LLMs for drafting or editing, and instead, fill the prose themselves. Fowler suggests that using LLMs for tasks like outlining or research is acceptable, but authors should remain mindful of the potential impact on their writing style. AI summary

Anthropic has filed its S-1 for an IPO, but details are still scarce, including the share price. Meanwhile, Google has released Gemini 4 Carbon, a new model that tops the Artificial Analysis v5 benchmark index, but it's not yet available for public use. AI summary

Ropes & Gray is developing AI for transactional diligence with OpenAI. Lawyers review grouped documents, summaries, and reports before advising clients.

Valve advised developers to focus on building a strong community and consistently updating their games to combat AI clone games, rather than relying on a single mechanic that can be easily copied. The company emphasized that trust built with players is a key barrier against clone games, and that continuous improvement is essential for small indie studios and established developers alike. AI summary

Improving document extraction accuracy with ai_extract and MLflow Prompt Registry involves using schema descriptions for field-level rules, global instructions for document-level logic, precision mode for structural complexity, and citations to diagnose errors. A repeatable workflow can be built using Genie Code, which automates the core of this workflow, scoring extraction field by field against labeled data, proposing schema and instruction fixes as reviewable diffs, and re-running the eval to show before-and-after results. AI summary

Anthropic's CEO Dario Amodei defends the company's actions, arguing that most of the criticisms levied against them are driven by a negativity bias and that they have taken significant steps to address safety concerns, despite the costs. He counters criticisms of the company's approach to AI development, arguing that their "race to the top" strategy is necessary to drive progress in the field and that many of their actions, such as delaying the release of models to ensure cyberdefenses are in place, were necessary to mitigate risks. AI summary

Proton is conducting "re-interviews" with its staff to assess who can continue working at the company, with the implication that the interviews may decide employment, following concerns about the use of AI in their jobs. AI summary

The GPT-6.1 Terra model is a fast and efficient text model for visual reasoning, precise computer use, and clear, natural writing, with a 1,050,000 context window and 128,000 max output tokens. Pricing is based on the number of tokens used, with text tokens costing $1.00 per million tokens and cached input tokens priced at 5% of the uncached input token rate. The model supports various endpoints, including Live, Chat Completions, Responses, Realtime, and Fine-tuning, with a maximum rate limit of 15,000 requests per minute (RPM) for the Grow tier. AI summary

Cunyan Ma, a Brown University student, has created an interactive map to visualize the hidden global labor powering AI, exposing the power structure that extracts low-wage labor from data workers in the Global South to make everyday AI algorithms function. The map links technology companies and their intermediaries to data workers around the world, aggregating existing data worker databases and news reports. By mapping the power structure, Ma aims to raise awareness about AI colonialism and promote a more equitable distribution of AI labor. AI summary

Here is a summary of the article in 3 plain sentences: The decline in tech jobs in the US and other countries cannot be attributed to AI, but rather to rising interest rates, which led to a hiring freeze and subsequent layoffs, particularly among junior workers. The recovery in tech hiring is characterized by a shift towards senior roles with AI skills, with postings for these roles increasing by 25% since 2025, while junior job postings have not recovered. The cost of the tech jobs bust is landing on the youngest workers, with unemployment rates among urban 15-to-29-year-olds in India and China reaching 18.9% and 14.5%, respectively, highlighting a generational imbalance in the labor market. AI summary