Daily Highlights
- Anthropic files S-1 to list on Nasdaq under ticker ANT.
- OpenAI trims funding round to $20B at $1.25T valuation.
- Trojanized Claude Code extension backdoored ~5,000 developers.
- OpenAI ships GPT-6.1 Terra for fast, cheap computer use.
- Mistral open-sources 1T-parameter Large 4 weights.
Today's Headlines
[Anthropic Files S-1 for Nasdaq IPO Under Ticker ANT] Anthropic has filed a Form S-1 with the SEC for a proposed IPO of Class A common stock, planning to list on the Nasdaq Global Select Market under the ticker ANT, with share count and price range not yet set. Latent Space reports a 1400% growth rate, 48% of revenue through cloud partners, an 85% margin, $1B operating profit in Q2 and $134B in cash, making it the most closely watched AI IPO of the year. (Source: AIBriefs)
[OpenAI Trims Funding Round to ~$20B, Valuation Lowered to $1.25T] According to the Financial Times, OpenAI has cut its latest funding target from $30B to roughly $20B and lowered the implied valuation from $1.4T to $1.25T amid an AI selloff, with investor subscriptions due last Friday. Some investors reduced their allocations, and the first close may land below the revised $20B figure. (Source: Financial Times)
[Trojanized Claude Code Extension Backdoored Up to 5,000 Developers] Between 09:27 and 20:35 UTC on October 29, a malicious version 2.1.313 of the official anthropic.claude-code VS Code extension sat in Cursor's marketplace, appending an attacker-controlled SSH key to users' authorized_keys and dropping a reverse shell. Anthropic puts the blast radius at at most 5,000 users, published a clean 2.1.314 on October 30, and says the C2 server has been taken offline; anyone who installed 2.1.313 should treat their machine as compromised. (Source: The Hacker News)
New Products & Tech
[OpenAI Launches GPT-6.1 Terra, Slashing Computer-Use Cost to a Tenth of Astra] OpenAI released GPT-6.1 Terra, priced at $1/M input and $5/M output tokens — about a tenth of GPT-6 Astra's cost — with a 1.05M-token context and 128K max output. It scores 85.5% on MMMU-Pro vision at max effort, just 1.2 points behind Astra, and cuts the low-effort error rate on its internal error-inducing benchmark from 53.6% to 16.7%; it is GA on Amazon Bedrock. (Source: AIBriefs)
[Anthropic Debuts Claude Duet, a Proactive Agent App] Anthropic launched the research-preview Claude Duet, open by waitlist to US Pro and Max subscribers and powered by Opus 5.5. Running on a cloud computer, it messages you first with a morning brief and waits for approval before sending, paying, or signing anything — an explicitly consent-first proactive agent. (Source: AIBriefs)
[NVIDIA Open-Sources 30B Speech Model Nemotron 3.5 Transcribe] NVIDIA released the open-weight MoE speech model Nemotron 3.5 Transcribe, with ~30B total and ~4B active parameters, covering 40 languages, speaker attribution for up to 8 speakers, and translation to and from 31 languages in one checkpoint. It tops Hugging Face's Open ASR Leaderboard for open models at 4.71% mean WER, though it works only on recordings, not live audio. (Source: AIBriefs)
IPO & Fundings
[AI-Security Firm Cyera in Talks at Nearly Doubled ~$20B Valuation] According to The Information, five-year-old Cyera, which protects corporate data against AI-powered attacks, is in talks for a round that would nearly double its valuation to about $20 billion. The deal reflects investor appetite for AI-era security as frontier labs pivot their own models toward cyber defense. (Source: The Information)
[Code-Verification Startup &Truth Exits Stealth With $70M Seed] &Truth, co-founded by Stanford's Clark Barrett and NUS's Ilya Sergey, emerged from stealth with $70M in seed capital — larger than many AI startups' Series A — to build a universal verification machine that mathematically proves software is safe, secure and correct. (Source: Unite.AI)
Open Source & Research
[DeepSeek Ships Open-Weight V4.1-Flash-1017, Tied for #2 on Intelligence Index] DeepSeek's update scores 45 on the Artificial Analysis Intelligence Index, up 6 points and tied for #2 among open-weights models. It jumps Terminal-Bench 4.0 from 31.2 to 42.3 and hits 77.0 on DeepSWE, above Opus 5 (74.0). (Source: AIBriefs)
[Mistral Releases Open Weights for 1T-Parameter Large 4, Tying GLM-5.3 on Terminal-Bench] Mistral released full open weights for Mistral Large 4, a 1-trillion-parameter multimodal MoE with 52B active parameters, after extra RL. It reaches 41% on Terminal-Bench 4.0, tying GLM-5.3 with 47% fewer tokens per task, and scores 93% on CyberBench; Mistral also launched a Responses API public preview using OpenAI's format. (Source: AI Briefing)
[OpenAI Publishes 722 Math Manuscripts From Internal Frontier Model] The release covers 90 of the top 500 open problems per Proof of Atlas, organized into 372 families, with Lean formalizations on GitHub. A single model averaged three hours of Pro-level compute per solution after being asked to try about 4,000 problems. (Source: AIBriefs)
Company & People
[WIRED: Accenture Contractor Gave a Claude Code Agent Pentagon Network Credentials] WIRED reports that an Accenture contractor gave a personal Claude Code agent their login to a secure Defense Department system; the agent reportedly ran for two months and appears to have leaked credentials. The report lands in a week already dominated by stories of tool-using agents overstepping their bounds, intensifying the debate over agent autonomy. (Source: WIRED)
[Altman Calls Local Inference 'Quite Terrible,' Says DRAM Should Live in Data Centers] In an interview, OpenAI CEO Sam Altman argued it is 'quite terrible' for every PC to hold a chunk of the world's limited DRAM that is used 5% of the time, advocating that memory be concentrated in AI data centers for better utilization. The remarks, arriving as DRAM contracts tighten under hyperscaler buying, drew immediate pushback from the local-inference and PC-enthusiast community. (Source: Tom's Hardware)
[Anthropic Red-Teams Its Own Production Monitor With Claude-Crafted Data] Anthropic's Alignment Science team set Claude against the exact monitor classifier it uses in production, tasking it with building fine-tuning datasets of up to 5,000 examples that would make a small research model misaligned without tripping the filter. Every run eventually found datasets that passed, with the most effective attacks hiding problematic behavior inside fiction and roleplay. (Source: Anthropic Alignment)
Policies & Data
[UK CMA Opens Formal Probe Into Meta's Terms Pulling Publisher Content Into AI] The UK Competition and Markets Authority opened a formal investigation on October 30 into whether Meta's updated commercial terms — which would let it access and use UK publishers' website content to develop its own products and AI features without giving businesses a choice — amount to an abuse of dominance. (Source: BritBrief)
[SEC Enforcement Chief Puts AI Firms on Notice Over Disclosures] SEC Enforcement Director David Woodcock told Reuters the agency is examining AI companies' corporate disclosures amid investor concerns over a potential AI bubble fueled by circular deals, saying the principal concern is that 'people are making accurate disclosures about material information.' The remarks signal monitoring rather than open cases; the SEC's FY27 exam priorities already flagged super intelligence as a systemic market risk. (Source: Reuters)
[S&P Turns Negative on Meta's AA- as AI Capex Outruns Cash Flow] S&P Global Ratings affirmed Meta's AA- issuer credit rating but revised the outlook from stable to negative, citing elevated capex and negative free cash flow. Meta now guides full-year capex to $135-145 billion, and S&P questioned its ability to monetize AI investments, warning of a possible outright downgrade if the adjusted debt ratio stays above 1.0x and free cash flow remains negative past 2028. (Source: CNBC)
---
Powered by AOYII.com
