A long day for OpenAI on every front — product, safety, politics, money — plus an open-weight model that worries Anthropic and an OAuth hole in the MCP SDK.
- OpenAI DevDay ships "Dots" agents, GPT-6.1 Sol, and a Pro 500 tier
- OpenAI pulls GPT-6.1 Astra over safety concerns
- The White House gathers 31 tech CEOs, and Congress counsels patience
- Anthropic flags Zhipu's GLM-5.3 as the most cyber-capable open-weight model yet
- An OAuth flaw in the official MCP Python SDK can leak tokens
- OpenAI is reportedly raising again, at a $1.4T valuation
OpenAI DevDay ships "Dots" agents, GPT-6.1 Sol, and a Pro 500 tier
At DevDay 2026, OpenAI introduced "Dots," a line of always-on agents, alongside a model it calls GPT-6.1 Sol and a new Pro 500 subscription tier.
The always-on framing is the part to watch: an agent that runs continuously is a different operational and cost profile than a chatbot you poke at, and it raises the same permissioning and oversight questions the rest of today's news keeps circling. The Pro 500 tier signals OpenAI is still segmenting upward on price rather than down. Both CNBC and the BBC covered the event, with the BBC noting the launch landed the same week a separate model rollout was delayed on safety grounds.
OpenAI pulls GPT-6.1 Astra over safety concerns
According to the BBC, OpenAI scrapped the rollout of GPT-6.1 Astra, citing safety concerns.
The timing is what makes it notable: shipping "Sol" and shelving "Astra" in the same week suggests the safety gate is being applied per-model rather than as a blanket pace, which is more reassuring than a company that ships everything or nothing. It also complicates the tidy DevDay narrative, and the BBC's coverage pairs the two events directly. Treat the reasoning as OpenAI's stated rationale, not independently verified detail.
The White House gathers 31 tech CEOs, and Congress counsels patience
France 24 reports that President Trump convened 31 tech CEOs to press for industry self-governance and to float rebranding the field around "super intelligence"; separately, CNBC quotes House Speaker Mike Johnson saying Congress should avoid a "knee-jerk reaction" on AI regulation.
Read together, both point the same way: the near-term regulatory posture in the US is self-governance and wait-and-see, not statute. For anyone building on these models, that means the binding constraints for now are vendor policies and liability, not a federal rulebook. The "super intelligence" rebrand is a framing move worth noting precisely because framing tends to precede policy.
Anthropic flags Zhipu's GLM-5.3 as the most cyber-capable open-weight model yet
Anthropic published research describing Zhipu's GLM-5.3 as the most cyber-capable open-weight model released to date.
Open weights mean the capability is portable and cannot be recalled, which is exactly why a cyber-capability assessment on an open-weight model reads differently than one on a hosted API. Anthropic is a competitor making this claim, so weigh it accordingly, but the underlying point — that frontier-adjacent capability is now shipping without an off switch — is the durable one. It reframes the day's US self-governance debate against a model no US policy can unship.
An OAuth flaw in the official MCP Python SDK can leak tokens
The Hacker News reports a flaw in the official MCP Python SDK's OAuth handling that can let a malicious server steal access tokens.
This is the unglamorous half of the agent story: as MCP becomes the plumbing connecting models to tools and data, its SDKs become high-value targets, and a token-theft bug in the reference implementation is about as load-bearing as bugs get. If you run MCP servers or connect to third-party ones, this is a patch-now item, not a read-later one. It also lands the same day OpenAI's "Dots" agents raise the question of what those agents are authorized to touch.
OpenAI is reportedly raising again, at a $1.4T valuation
Per Bloomberg, OpenAI is in early talks for a new funding round of roughly $30 billion at a valuation near $1.4 trillion.
These are early-talks figures reported by Bloomberg, not a closed round, so treat the numbers as reported rather than final. Still, the scale sets the backdrop for everything above: the pace of launches, the price segmentation, and the appetite for self-governance all make more sense next to a capital requirement this large. A company raising at this level has strong incentives to keep shipping and to keep the regulatory environment light.
Tools worth your time
- GPT-6.1 Sol (OpenAI) — the model OpenAI did ship at DevDay, aimed at developers and heavy users evaluating the next tier of general-purpose capability. openai.com
- Dots (OpenAI) — always-on agents for people who want tasks running continuously rather than prompt-by-prompt; worth a look if you're already building agent workflows and can scope their permissions tightly. openai.com
- GLM-5.3 (Zhipu AI) — an open-weight model notable enough that Anthropic wrote it up; relevant to researchers and teams wanting frontier-adjacent capability they can run and inspect themselves. z.ai
This piece was drafted autonomously and reviewed by a human editor before publishing. Every machine-written article on Nerdic carries this mark.
