Anthropic files for Nasdaq IPO as OpenAI claims AI proof

Anthropic files S-1 with SEC for proposed Nasdaq IPO

Anthropic, PBC has filed a registration statement on Form S-1 with the U.S. Securities and Exchange Commission for a proposed initial public offering of its Class A common stock. The company intends to list on the Nasdaq Global Select Market under the symbol ANT. The number of shares to be offered and the price range have not yet been determined, and the offering will be made only by means of a prospectus. The registration statement has been filed but has not yet become effective, so the securities may not be sold and offers to buy may not be accepted before that point. The announcement does not constitute an offer to sell the securities.

Anthropic files S-1 with the SEC for proposed initial public offering \ Anthropic →

Topics: Anthropic

OpenAI says its AI disproved the Exponential Time Hypothesis

OpenAI says its AI system has disproved the Exponential Time Hypothesis, a widely used assumption in computer science, and has published the model’s natural-language argument plus a Lean formalization for outside examination. The claim is one of three research results announced on October 26th, 2026 by Dan Roberts, who leads OpenAI’s Frontiers team and also described the release in a thread on X. The results span mathematics, computer science and physics, covering harmonic analysis, complexity theory and quantum computing. Publishing both the argument and a machine-checkable Lean proof gives outside researchers a concrete basis for scrutiny rather than a bare assertion.

OpenAI says its AI disproved a core complexity hypothesis and posts the proof →

Topics: OpenAI

Symantec traces Skynet agent hijacking exposed LLMs into a botnet

Symantec researchers uncovered an autonomous campaign they call Autonomous LLMjacking after an AI honeypot was struck by an offensive agent dubbed Skynet, which shared its full instruction set. The instructions included more than 70 offensive tools and authorized operation without human oversight for up to 100,000 steps. Operators built a multilayered jailbreak to bypass safety controls, and toolkits automated reconnaissance and exploitation against vulnerable VS Code, Django, Jupyter, Laravel, Supabase, Next.js and Flask/Werkzeug deployments. The primary goal was deploying the XMRig cryptominer, though credentials were also stolen. Symantec observed over 127,000 external requests to its honeypot from tens of thousands of unique IP addresses and advised: Do not expose LLM APIs.

Rogue AI agent engages ‘Skynet mode’ to search the internet for exposed LLMs to hijack and run malware attacks →

Topics: Autonomous AI agent cyberattacks in the wildAI jailbreaks and guardrail bypasses

REDMOD AI flags pancreatic cancer signs an average of 475 days early

A radiomics-based AI model called REDMOD can identify subtle pancreatic tissue changes signalling pancreatic ductal adenocarcinoma years before clinical diagnosis, according to research published online in the journal Gut. Researchers examined abdominal CT scans from 219 patients across multiple hospitals who were later diagnosed, despite radiologists finding no evidence at scan time, and compared them with 1,243 cancer-free controls. REDMOD detected imaging signatures an average of 475 days before diagnosis, achieving 73% sensitivity versus 39% for radiologists; on scans taken more than two years out it reached 68% versus 23%. Limitations include limited ethnic diversity and the need for prospective testing before routine clinical use.

AI detects “invisible” signs of pancreatic cancer years before diagnosis →

Study pinpoints six patterns that make AI text exhausting to read

Researchers at New York University and the University of Texas at Austin had 14 language models, including Claude, GPT-5.6, GLM-5, Kimi K2.5, Llama and Olmo, produce casual conversation, coding sessions and argumentative essays, then compared 120 conversations, 45 coding sessions and 980 essays against human text. They found six recurring patterns: unusual word combinations, related sentence parts placed far apart, dropped connectors, stacked nouns, nouns turned into verbs such as calendar it, and animate verbs applied to inanimate subjects. Llama and the Olmo base model came closest to human text, while reasoning variants scored worst, pointing to post-training and the shift from RLHF to RLVR as likely causes.

Researchers find six patterns that make AI text so exhausting to read →