[ The Morning Build ]
3 min read5 storiesAIBig TechStartups

The Morning Build for November 3, 2026: OpenAI Unifies Chat and Work, Squarespace Cuts 19%, and Weight-Space Editing by Astra

Today’s stories link product and policy responses to AI’s spread: OpenAI is merging Chat and Work in ChatGPT and aligning token allowances; Squarespace is restructuring with a 19% workforce reduction as it pushes into AI-native features; researchers report Astra reading and writing model weights; OpenAI raised bug-bounty payouts and launched a security sprint; and Anthropic updated Claude Code to block game cloning.

OpenAI merges ChatGPT’s Chat and Work tabs, unifying files, plugins, and token allowances

  • What happened: OpenAI told users it will merge the Chat and Work tabs inside ChatGPT, initially for select users to gather feedback before a wider rollout; Codex remains a separate app section. The company said ChatGPT and Codex will draw tokens from the same allowance pool and that some users may see more frequent resets depending on usage and model.
  • Why it matters: Engineers and product teams will have a single workflow surface for exploratory prompts, files, plugins, and web actions inside ChatGPT, and token accounting will now span ChatGPT and Codex, which can change quota behavior for some users and integrations.
  • Outlook: Select-user feedback period precedes a wider rollout, with the company collecting input from early access users before expanding the merged experience.

Sources: techcrunch.com

Squarespace restructures and cuts about 19% of staff, roughly 300 roles

  • What happened: Squarespace confirmed a global workforce reduction of approximately 19%, representing around 300 employees, calling the move a restructuring to simplify operations and sharpen priorities as it prepares for its next phase of growth.
  • Why it matters: The layoffs reduce headcount at a major web-hosting and site-builder platform as the company accelerates AI initiatives like Blueprint and the upcoming Forma product; that could shift product staffing, timelines, and partner integration availability.
  • Outlook: Squarespace’s upcoming AI-native platform, Forma, including its waitlist and rollout milestones, will show how the company reallocates resources after the restructuring.

Sources: techcrunch.com

Astra demonstrates reading and writing neural network weights to fix and generate models

  • What happened: Researchers reported that GPT-6 Astra can analyze raw model weights to identify a network’s task, edit weights to fix failures, and generate small networks from scratch; examples include raising MATH-500 from 7.9% to 48.1% by editing 0.016% of Qwen3.5-0.8B weights and producing one-shot MinAtar agents that outperform PPO trained for 100 million steps.
  • Why it matters: The results show an LLM performing direct weight-space reasoning and edits without gradient-based optimization, which implies a new vector for model debugging, targeted fixes, and zero-shot network generation that could change workflows for optimization and interpretability research.
  • Outlook: The next experiments will push Astra to tool-assisted and larger-scale scenarios, including editing at the 1.5 million parameter scale with a Python sandbox and limited forward passes.

Sources: rikiyat.github.io

OpenAI raises bug-bounty payouts to as much as $172,000 and launches a security sprint with Cathedral

  • What happened: OpenAI increased bounty payout tiers so critical product vulnerabilities can earn up to $30,000, exceptional findings up to $100,000, and with loyalty and campaign bonuses payouts can reach roughly $172,000; the company also announced a security sprint partnering with Cathedral and said prior sprints involved 25% of product engineers.
  • Why it matters: Higher monetary incentives and a coordinated sprint with an external security firm change the economic calculus for external researchers and may surface more high-severity findings, which affects risk assessment and vulnerability timelines for products like ChatGPT and Codex.
  • Outlook: Time-limited campaigns with boosted payouts around new feature launches and the Cathedral-hosted security sprint are the immediate signals that will produce vulnerability disclosures and retroactive payout adjustments.

Sources: the-decoder.com

Anthropic updates Claude Code to block cloning existing video games from system prompt

  • What happened: Anthropic added two sentences to Claude Code’s Opus 5.5 system prompt to prevent the tool from recreating other people’s games; the rule blocks clones even when renamed, while still allowing users to build original games, genre titles, and projects with custom assets.
  • Why it matters: The policy change constrains automated code generation that reproduces identifiable game IP, affecting developers who used Claude Code to generate ports or derivative works and signaling tighter on-model enforcement of intellectual-property rules in coding assistants.
  • Outlook: Anthropic said it plans to keep tweaking the rule going forward, making further Opus 5.5 policy adjustments the next concrete milestone.

Sources: the-decoder.com