AI's legal and competitive fault lines are showing. A jury found Arm breached its contract with Qualcomm but awarded just $2, and federal prosecutors named the OpenAI engineer accused of bugging an executive meeting room. Meanwhile Anthropic pulled forward its timeline for automated AI research after fresh test results.

Summary: A Delaware jury found that Arm breached its contract with Qualcomm and interfered with Qualcomm's partnership with Meta. The award was a symbolic $2. Qualcomm had accused Arm of withholding software and design tools and leaking a licence-termination threat to damage the Meta deal. The jury rejected claims about Arm's pricing promises. Qualcomm's existing licence runs through 2033.

Why it matters: Arm licenses the blueprints behind most phone chips, so disputes over its terms ripple across the entire mobile and AI hardware supply chain.

Source: CNBC

Summary: Federal prosecutors charged OpenAI engineer Abdulrahman Alfozan with secretly recording company leaders using a device hidden behind a TV in the San Francisco office. Investigators found the bug on 30 September, two days after it was allegedly placed. A linked chatbot account showed searches on remote listening, encryption and disguising a homemade recorder. OpenAI says it ran its own models on forensic copies of the device to identify him.

Why it matters: US authorities are treating the case as a national security issue, a sign of how sensitive frontier AI work has become.

Summary: Anthropic co-founder Jack Clark raised the odds of fully automated AI research by 2027, up from his earlier 2028 estimate. The shift followed a test called TasteVal from P-Zero Research, which measures how well AI systems plan experiments for building AI. Anthropic's Opus 5.5 performed strongly at a fraction of the cost of human expert teams. Clark had previously doubted whether AI could make the creative leaps needed for self-improvement.

Why it matters: If AI can design its own experiments cheaply, the pace of AI development could accelerate well ahead of current expectations.

Summary: Google engineers let Gemini 4 Argon run unsupervised on a large codebase. It proposed six speed-ups, but only two were deemed viable. A second round with review guidelines and full tests showed how much direction matters. The lesson is to treat AI suggestions as creative exploration rather than finished answers.

Why it is interesting: It is a concrete reminder that AI output is raw material, not a drop-in solution.

Summary: A guide walks through setting up a dedicated email inbox for an AI assistant so it can process updates and write short briefs. Steps include requiring approval before any email is sent, assigning an inbox address, and testing with sample messages. It warns that an inbox does not remove previous access permissions and suggests confining early trials to fictional scenarios.

Why it is interesting: It shows how quickly everyday email workflows could be handed to agents, and where the safety traps sit.

Source: Fasrad

Summary: Microsoft patched over 1,400 security flaws in a single month and attributed many of the discoveries to AI tools. The volume shows how AI is being pointed at defensive work, not just new features.

Why it is interesting: The same technology raising security fears is now helping find the holes faster than humans alone could.

Context: Based on the coding case study where AI proposed six fixes but only two were viable, this prompt turns AI from an automated solver into a disciplined exploration tool.

Prompt: You are reviewing [codebase or document]. Propose up to six distinct improvements to [specific goal, e.g. speed, clarity, cost]. For each one, state the expected benefit, the risk, and how I could test it before adopting it. Rank them and tell me which two you would implement first and why.

Example use case: A team asks the model for performance fixes on a slow service, then uses the ranked shortlist and test notes to implement only the two safest, highest-impact changes.

Summary: Anthropic told prospective investors that its third-quarter revenue rose 45% from the previous quarter to $16.8 billion, according to two people who attended its investor day in San Francisco. That is up from $11.6 billion in the second quarter and $4.7 billion in the first. The company also said it was profitable on an adjusted basis for the quarter. Anthropic has not commented, and its stock market listing is expected after the November US midterm elections.

Why it is important: Revenue more than tripled in six months, which shows how fast businesses are now paying for AI tools. The figures will also shape how investors value OpenAI and other AI labs heading for a listing.

Source: Reuters

Summary: Nvidia CEO Jensen Huang announced roughly €2 billion in upcoming European AI initiatives. Separately, Chinese tech firms are reported to be buying gaming and workstation GPUs to sidestep export restrictions on AI chips.

Why it is important: Both moves show the AI hardware map being redrawn by regional investment and export controls.

Source: Bloomberg

That's it for today's Daily Pulse. Forward this to one person who wants to stay ahead of AI. See you in the next one. - Nicolas