OpenAI, Google, Microsoft and Anthropic backed a single robots.txt signal that lets websites opt out of AI training, while OpenAI detailed its system for monitoring model misalignment and tightened weapons information for teen users. On the business side, Anthropic is expected to get only a junk credit rating after its IPO, and Alphabet had its worst trading day on record amid worries about AI spending. The subscription race for coding agents also heated up: OpenAI doubled Pro allowances and Anthropic dropped the 5-hour cap on its Ultra plan.
On October 30, Google committed to support the “__usage_ai-training” signal (an IETF draft proposal) from early 2027, and Microsoft, OpenAI and Anthropic committed the same day. Sites need one line instead of blocking each crawler, while company-specific controls such as Google-Extended keep working – a meaningful step for publishers.
An October 29 post describes a chain of activation probes, a reviewer model, investigator agents and network-log cross-checks, covering all RL runs of models at least as capable as GPT-6 Astra and all internal Codex use. OpenAI says a replay would have alerted more than 30 hours before the Hugging Face breach, but admits gaps such as no round-the-clock paging for most traffic.
On October 30, OpenAI said ChatGPT will give less detail on past mass attacks for all users, further restrict weapons and security-system information for teens, and may automatically end chats showing fixation on violence. It will publish a Community Safety Report every six months from early 2027; it made more than 200 law-enforcement referrals in Q3 alone.
The new version (October 29) adds techniques such as manipulating how an agent decomposes a goal so a restricted action looks necessary, or blocking its usual route to force an attacker-chosen fallback plan. The core message: agent security must assess the whole plan, since each individual step can look benign.
From October 30, the $100/month plan is called Max and the $200/month plan Ultra, with prices unchanged; Ultra drops the 5-hour session limit for verified accounts (weekly limits unchanged) and includes $200 in monthly API credits. The move squarely targets heavy users of coding agents such as Claude Code.
After user backlash, OpenAI (October 29) doubled Work and Codex allowances for Pro 100 and Pro 500 and moved all Pro 200 accounts to the higher allowance; subscription prices are unchanged. Usage allowances for coding agents are becoming a main front in the OpenAI–Anthropic competition.
On October 29, Tencent Cloud said WorkBuddy has handled more than 9 billion task requests across 50+ industries and is being expanded for international enterprises; its Hy4 model (first with visual understanding, stronger on long-horizon tasks) is coming to general release, alongside a cheaper Hy4 Flash and the TokenHub single-API model platform.
According to the FT (reported by TechCrunch on October 30), Blackstone – anchor investor in a $63 billion chip-financing deal for Anthropic – now expects a BB rating rather than the investment grade it previously predicted. That could raise borrowing costs as Anthropic needs huge sums for compute, just after filing to go public.
Alphabet, Amazon, Meta and Microsoft plan more than $740 billion in capex this year; Alphabet’s Q3 free cash flow was -$9.1 billion and Meta’s -$2.1 billion. Alphabet shares fell 12% (about $500 billion in market value) on Thursday and the Nasdaq dropped 2.9% on Friday, signaling investor doubts about AI infrastructure returns.
On its Q3 earnings call (October 30), DoorDash said it has received only a few hundred orders from agents such as Muse, Instinct and Fo, versus about 3 million deliveries a day. Agentic commerce remains very early, though DoorDash stays open to partnerships and already offers its own MCP connector.
According to Thanh Niên (October 30), Teradata introduced its Autonomous Knowledge Platform, with a Tera Context Engine that connects internal data to agents without moving it, plus Teradata Factory for running AI on-premises or across clouds – relevant to data-residency rules. The aim is to help companies move from GenAI pilots to running agents in supply chain, risk management and customer care.
If these digests help you make sense of technology, a small contribution keeps NHỊP picking out what is worth reading.
Scan the QR code with a Vietnamese banking app.
Download QR ↓ · View full image ↗