Claude's overnight game rule, Claude plan renames, and Jira's agent dashboards

Atlassian on agent sprawl — Atlassian's AI-at-work team argues that unmanaged agents acting with your data and permissions are an accountability risk, not a shadow-IT problem, and that the fix is defining what agents may do, on which data, with what review — plus enough visibility to clean them up. A workable shape for the agent policy you keep postponing: split low-risk actions (search, read, draft) from high-risk ones (deleting content, permission changes, regulated data) and let teams build inside those lines. DocuSign reports ticket drafting cut from about 20 minutes to 5 minutes per issue under governed agents, with one QA team reclaiming over 10 hours a week. Atlassian (AI at Work)

Jira Autumn 2026 release — Jira's latest release gives agents their own identity on boards and work items, one admin view of where every agent is embedded, and an agent usage dashboard that ties agent activity to cycle times, throughput and shipped work. If your team runs work through Jira, this is the first built-in answer to whether agents are actually moving delivery or just generating activity — no extra tooling to buy. The dashboard needs no configuration, boards in company-managed spaces load 22% faster, and the whole release lands across Jira Cloud during November. Atlassian (Inside Atlassian)

Claude Ultra plan — Anthropic's help pages confirm the Max 20x plan is now called Ultra: ten times the Pro plan's weekly usage for $200 a month, with no five-hour session cap once the account passes Anthropic's automatic verification checks. Anyone on your team with a Max 20x subscription has already been moved over at the same price; the session cap they may have grumbled about disappears only after verification, and a temporary limit applies until then. Weekly caps remain across all models, with a separate weekly limit for the Fable models. Claude Help Center

Claude Fable 5 — Anthropic has spelled out how the Fable models are billed: on Max, Ultra and premium Team seats they're included, taking up to half the weekly limit; on Pro and standard Team seats they're excluded and run on pay-as-you-go credits. On standard Team seats — probably most of yours — a developer switching to Fable creates charges beyond the subscription; give whoever owns the budget a heads-up. Claude Help Center

OpenAI's round — The Wall Street Journal says OpenAI has about $11bn committed toward a round now capped at $20bn, down from the $30bn first sought, with interest from some large investors described as tepid. More fuel for the AI-spending scrutiny your CFO is already voicing — expect tougher questions about your own agent budget. OpenAI plans to list as early as next year, while Anthropic's IPO is expected mid-November. Trending Topics

robots.txt policy token — OpenAI, Google, Microsoft and Anthropic say they will honor one shared robots.txt signal, proposed at the IETF, letting a site opt out of AI training across all of them at once instead of crawler by crawler. If your product has a public site, your web team can state the preference once — but honoring only starts in the coming months or next year. Anthropic already reads the token; OpenAI says it will within months, and Microsoft notes opted-out sites still appear in Bing. The Verge

ChatGPT share-link bug — Security firm Parameter showed a malicious public ChatGPT share link could run attacker code inside the signed-in session of anyone who opened it, enumerate their private conversations and send the contents to an outside server. OpenAI rated the report its most serious category — and the takeaway for your team stands: treat AI share links from strangers like attachments from strangers. Opening the link was the only action the victim needed to take. Parameter

Cathedral's Anthropic test — During an authorized security test, startup Cathedral used AI agents to reach sensitive internal Anthropic systems — HR, sales and IT records, including salary details — in under 12 hours. Anthropic says no customer chats, source code or model weights were touched, and accuses Cathedral of overstating the risk and breaking the test programme's rules; either way, third-party SaaS is the attack surface your agents inherit too. India Today

OpenAI's funding round — OpenAI has secured about $11bn toward a round it first pitched at $30bn, and has cut the target to $20bn after what the Wall Street Journal called a tepid response from some large investors. Backers from the UAE, including MGX, are expected to lead. A wobble in the financing behind the biggest vendor roadmap you depend on, with the first close likely below even the new target. The round values OpenAI at $1.25tn after the new money. TechCrunch