Created automatically with AI: the texts and images in this edition are AI-generated, based on the linked original sources. Errors may occur.
Google is pausing its open-source rewards program because of a flood of invalid AI reports. A new AI task force is being established in the United States; there is also research on self-optimizing agents and an open lunar model from NASA and IBM.
AI-generated image
01TechCrunch
Google temporarily halts open-source rewards program because of invalid AI reports
Google suspended its program rewarding the discovery of security vulnerabilities in open-source software as of October 1. The company said the number of automated submissions had risen significantly, and the vast majority were invalid. Google plans to provide an update on the program in the first quarter of 2027. Until then, the company is directing participants to other bug bounty programs.
Why it matters
AI-generated false reports can make it harder to review genuine security vulnerabilities and tie up the resources of researchers and maintainers. Google's pause shows how automated submissions can burden existing reporting processes.
Trump establishes “Super Intelligence Force” led by US intelligence coordinator
US President Donald Trump has announced a “Super Intelligence Force” to coordinate government activities related to artificial intelligence. Intelligence coordinator Jay Clayton will chair it; other officials include FTC head Andrew Ferguson and Pentagon representative Emil Michael, according to reports. The task force is to assess AI's opportunities and risks within 120 days. Trump uses “Super Intelligence” as a term for AI; here, the term does not mean actual superhuman intelligence.
Why it matters
The task force closely links AI policy with national security and the goal of remaining a leader in international competition. According to heise, no representatives of civil rights or scientific organizations have so far been named to the panel.
Google research aims to protect self-optimizing AI agents from overfitting
A research team from Google Cloud AI Research presents RRSI, a method for optimizing the working environment of AI agents. It is intended to prevent agents from merely memorizing their test tasks. In the reported tests, scores on previously unseen benchmarks increased by up to 4.7 points, while token use fell by about 30 percent compared with a non-regularized variant. The underlying language model remained unchanged.
Why it matters
The work examines how improvements to agents can transfer to unfamiliar tasks rather than merely optimize performance on known tests. The results come from benchmarks and do not automatically demonstrate comparable effects in practical use.
NASA and IBM release open AI foundation model for lunar research
NASA and IBM have introduced an openly available model for analyzing lunar data. It was trained on image and measurement data from several missions and is intended to help, among other things, with detecting craters and predicting possible ice deposits. According to a technical report, it performed at least as well as comparison models on several tasks; IBM reports a lower error for predicting ice deposits. The results depend on the tests in question.
Why it matters
The model brings together different types of observational data and could help researchers tackle tasks with few labeled examples. The authors see it as a basis for further analysis, not a replacement for physical measurement instruments.
Aleph Alpha reports political biases in tested Chinese AI models
An investigation developed by Aleph Alpha tested models from Alibaba, DeepSeek, and Moonshot AI on 967 politically sensitive topics. Under the company's evaluation system, only 17 to 41 percent of responses were considered balanced; other responses reportedly echoed government positions, evaded the question, or refused to answer. Aleph Alpha itself has a business interest in differentiating its own “sovereign AI” offerings from Chinese competitors. The results should therefore be read in light of these potential conflicts of interest.
Why it matters
The report raises questions about how political perspectives in AI responses are measured and evaluated. The investigation comes from a provider that competes with the companies tested in this market.
Automatically summarized with AI and checked against the source texts retrieved. Original sources are linked for every story; errors may still occur. An editorial overview from Medienfuchs.
