Week in review: Grok Bot Pages, and phone apps still rolling out

Week in review: Oct 19 to 23#

1. Pages: ask your Bot for a page, then have it keep the page current (Grok Bot; confirmed)#

The week's top feature is now in the official notes. Grok Bot's changelog entry for v0.75.2, dated Oct 22, says: "Bots can build Pages, such as reports, memos, and dashboards, that you edit together or have the Bot keep up to date." It adds that "Page owners can share a Page with their whole organization or with another team in it," and that "The dashboard's Pages section lists the Pages you've shared, where you can unshare or delete them."

@bot announced Pages on Wednesday: "Ask Bot for a page, then edit it together or have Bot keep it up to date." Later that day, Jenny Wen, whose X bio reads "head of design @cursor_ai @bot," wrote at 5:07 PM ET that "my bot makes pages for these and updates them on a weekly basis," using routines. The changelog still doesn't say how a Page refreshes. Our read: ask for the page first, then give a routine the job of updating it. (launch story · night-one uses)

2. Phone apps: announced and rolling out, not yet in release notes (Grok Bot; announced)#

@bot posted at 6:14 PM ET Thursday: "Grok Bot can now use Calendar, Reminders, Contacts, and more apps on your phone. Rolling out now on iOS and Android." A follow-up says: "Update to the latest version to try it." Release notes and docs don't mention it yet. At 6:29 AM ET Friday, the App Store still showed 1.17.0 from Oct 21 ("Improved design of Connect Apps screen"), Google Play still showed an Oct 18 update, and the v0.75.2 changelog entry has no phone apps item. Update the app and expect it to reach you over time. (our story)

Evan Bacon, whose X bio reads "Frontend lead for Grok at @SpaceXSI," wrote at 6:33 PM ET Thursday: "I've been using this to plan out my day during my commute so I have a clear schedule before I get to work." (staff post) Three users report a limit (user reports): Paul Lechner wrote at 10:33 PM ET Thursday that the Reminders tools "only worked for me while the Grok Bot app was open on my iPhone"; A. Sharma wrote at 10:42 PM ET that "my bots can't add a reminder unless the app is open on my phone"; and Juan Gonzalez wrote at 12:29 AM ET Friday: "I need to keep the app open so it can read the calendar". We found no staff answer to these in the replies we checked. Our read: if a routine needs your phone, keep the Grok Bot app open when it runs.

The rest of the week#

• Grok Bot: 1Password sign-in for your Bot, rolling out "over the next few days" (read); X Spaces, Lists and livestreams through the X connector (read); the 0.71.2 notes with 20 files per message (read); the iPhone Action Button (read); and users adding a phone line through a calling service (read).
• Grok Bot known issues: Bot computer updates on Oct 17, 21 and 22 dropped installed apps (read), and Bots went missing after the X Subscription switch; staff say don't reset or pay again (read).
• Cursor: Bugbot on Bitbucket and GitLab-triggered automations stalled from Oct 21 (read).
• Anthropic: API version 2026-10-21 drops temperature and new SDKs switch to it by default (read); the Claude Duet waitlist (read); continuous scanning in Claude Security (read); Claude Code 2.1.304 to 2.1.307 (read, read).
• OpenAI: approval for routers and AI gateways from Nov 20 (read); the Decisions API is generally available (read); 12-hour prompt caching (read); PIN-locked ChatGPT chats (read); refunds for credits drawn in the OpenCode attack (read).
• Hark: Adcock says one day can use a fifth of your month (read), users report videos made by chat (read), and crypto guides now name Hark for Binance's wallet skill (read).

Grok BotWhat's new#

Update, Oct 23, 7:07 AM ET: Grok Bot 0.75.2, the release behind Pages, lets Bots make videos and pay with Shop Pay. The changelog says "Bots can create videos from a description or from an image you share" and "Bots can pay with Shop Pay, through Shop checkout where a store accepts it and with a one-time Shop card elsewhere." It also adds one-tap X Money and memory shared across a personal Bot's chats. The notes give no plan, region or price limits, and we haven't tested a purchase.

Most Claude apps start new sessions in auto mode next week (Anthropic; confirmed)#

If you're on Claude Pro, Max or Team, new sessions will stop asking you to approve each step, starting next week. @claudeai posted at 4:04 PM ET Thursday: "Starting next week, sessions across Claude web, desktop, and mobile apps will run in auto mode by default on Pro, Max, and Team plans." Anthropic's article says "most Claude apps" and gives no day. Our story covers how to keep Manual.

Codex CLI 0.164 saves prompts as presets (OpenAI; confirmed)#

Update Codex CLI to 0.164.0 to save prompts you reuse. The release notes say: "Save reusable prompts and slash commands in ten numbered /presets slots, with a keyboard menu for quick access." /model and /permissions now take arguments too.

Claude API run by Anthropic, now through Google Cloud (Anthropic; confirmed)#

Google Cloud customers can call the Claude API that Anthropic operates, Managed Agents included, and pay on their Google Cloud bill. Anthropic's release notes say it's "the Claude API operated by Anthropic and reached through your Google Cloud project." It creates a new Anthropic organization, and API keys don't carry over.

Open Responses moves to the Agentic AI Foundation (OpenAI; confirmed)#

OpenAI contributed Open Responses, its Responses-style request format, to the Agentic AI Foundation on Oct 22. The Linux Foundation says it lets developers "describe requests and outputs once and run them across model providers with minimal translation." Test a provider against the acceptance tests before you switch.

What METR and Redwood found in OpenAI's agent monitoring (OpenAI; explainer)#

Two outside researchers tested how OpenAI watched its internal agents as of July 21. METR says OpenAI "has stated that all of the vulnerabilities we identified that impact their production safeguards stack have been patched," and adds: "We have not evaluated these claims." Our explainer turns their design ideals into a checklist for your own agents.

Crypto guides tell users to run Binance's wallet skill in Hark (Hark; user reports)#

Since Thursday afternoon, posts on X have named Hark's free chat as the agent to drive a Binance Agentic Wallet for NFT mints. We found no tie between Hark and Binance. If you connected one, check its limits in the Binance App, review approvals and sign the agent out.

Opinion · Friday wrap#

This week's column: Anthropic now shows platforms when it restricts one of their end customers, but the webhook event isn't sent yet, so our top pick is a small poller. Second, map every route to OpenAI before the Nov 20 approval rule. Third, a sign-in check for Hark. (read the column)

Grok BotThe digest#

Grok Bot#

• "All screens on the shared computer are in use" can be a stuck reservation. Colin said at 11:59 AM ET Thursday it's "a stuck screen reservation on our side, not something you did," and that "Restarting the app, idling, or closing screens from the other bot can’t release it." It's "a known issue the team is already tracking." In that case staff refreshed the computer from a backup, without a Reset. (Grok Bot; confirmed, staff; tracked)
• Every command stalls on the Bot's computer after a scheduled update. For one user since about 9:00 AM ET Oct 21, Colin said commands sit "for about 14 minutes and then fail, while bots that only use connections like Gmail, Calendar and Trello keep working." It's a tracked known issue. His advice: "please don’t reset again, since it hasn’t helped and interrupts your bots each time." Related: our Bot computer updates story. (Grok Bot; confirmed, staff; tracked)
• No Public profile switch after the move to the Grok dashboard. Colin said "there is currently no switch on the web for making your @handle page public." It's a tracked known issue, and the team is looking at bringing it into the new settings, "though I can’t give a date." (Grok Bot; confirmed, staff; tracked)
• Non-web ports still only work on port 443 for some hostnames. On Bot computers, hostnames that resolve to 198.18.0.1 "only work on port 443 right now," Colin said at 1:43 PM ET Thursday. "The team is actively working on it." Other non-web traffic leaves from "a small set of shared cloud addresses," and "each new connection can come from a different one," so check your server's firewall if connections by IP fail. (Grok Bot; confirmed, staff)
• Teams: mention the Bot, not Cursor. When a teammate whose Teams account isn't linked @mentions the Cursor app, Colin said, "Cursor replies with a “connect your account” prompt and your routine doesn’t start." His advice: "ask teammates to address the bot by @mentioning its own Teams account or by name, without @mentioning Cursor." (Grok Bot; confirmed, staff)

Cursor#

• The Cursor Grok 4.6 usage-bar label is updated. deanrie said at 3:03 PM ET Thursday: "we’ve updated the label display in the usage bar for Cursor Grok 4.6 on our side. You don’t need to update the app for this." Earlier in the thread he said the "included API usage" label was "just a display issue" and that Grok 4.6 bills from the Cursor Models pool shared with Composer and Auto. Check Dashboard > Usage on cursor.com for the breakdown. (Cursor; confirmed, staff)
• /review-bugbot "Unknown model ID" error: an update "should resolve it." kevinn said at 1:38 PM ET Thursday the error "wasn’t caused by anything in your setup" and "An update went out this morning that should resolve it." Fully quit and reopen Cursor, start a new chat, and run /review-bugbot again. Staff haven't called it fixed. (Cursor; confirmed, staff)
• Agents Window stuck on "Planning next moves." kevinn said the agent had finished, but the window "stopped picking up the agent’s updates." It's a tracked known issue "caused on our side, so updating Cursor isn’t expected to change it." Run Developer: Reload Window, or open the agent at cursor.com/agents. (Cursor; confirmed, staff; tracked)
• Orange needs-input dot stays on finished chats. Colin said the Agents list is "holding on to a saved “waiting on you” state," mostly on chats run as a task for another chat, and he has "flagged this to the team." Open the chat and click another, or archive it. (Cursor; confirmed, staff; flagged)
• File list links break on network-share workspaces. Colin said the link "rebuilds the file location without the server part of the network path." He has "flagged this to the team." Click the file name on the edit card instead, or map the share to a drive letter. (Cursor; confirmed, staff; flagged)

Hark#

• Adcock pitches Hark to Fortune's editor-in-chief. Alyson Shontell (@ajs), whose X bio says she is Editor-in-Chief of Fortune, wrote at 1:08 PM ET Wednesday that after trying two other assistants, Instinct and Muse, "I think I like Muse better??" Brett Adcock, Hark's founder (Business Wire), replied at 11:56 AM ET Thursday: "Give Hark a try, it’s better at proactivity". That evening, at 6:26 PM ET, Hark's own account quoted a user who called Hark "insanely thoughtful" (JeffButcher, user report) and wrote: "Don't just take our word for it, or Jeff's word for it. Try it out". (company posts)
• Usage limits again; Adcock suggests upgrading. A user in that thread wrote at 4:08 PM ET Thursday: "Hark is better but no usage resets means that I am stuck for 10+ days now." Adcock replied four minutes later: "Good to hear. What plan are you on. Why not upgrade?" Haseeb Mir, whose bio says he is a software developer, posted at 1:38 AM ET Friday that "the usage limits run out way too fast when you actually use it as a daily agent," asking for "weekly and monthly resets on the plans." Hark hasn't announced a change since Adcock said on Monday it would "re-look" at limits (X; our Oct 21 story). (user reports)
• iOS ratings jump as the app stays on 2.4.0. Apple's lookup data for the App Store listing showed 4.55 stars from 1,060 ratings at 6:14 AM ET Friday, up from 3.74 from 337 when we checked at 6:09 AM ET Thursday (Thursday's issue). Our math: 723 more ratings in about a day. The listing page itself showed 4.6 from 1,265 at 6:14 AM. The version is still 2.4.0, released Oct 16, so the jump didn't come with an app update, and Hark hasn't said what drove it. (store listings)

Anthropic#

• Claude proved 239 OSS-Fuzz entry points crash-free, under stated assumptions. Anthropic's Frontier Red Team and Theorem report that "Over the span of a few weeks, Claude successfully patched the bugs it found and then proved that there are no crashes on any input (up to 64 MB) for 239 entry points across 123 different OSS-Fuzz libraries." "This means these proofs currently assume that the program runs as a single-threaded static binary with a fixed memory budget on a RISC-V CPU using Linux OS." Its own caveat: "“Formally verified” does not mean “bug-free.”" It also says the proofs "can take hundreds of millions of tokens to produce." Our read: to try it, start with Theorem's Claude skill and the proofs. (confirmed)
• Anthropic donates the Agent Skills standard to the Agentic AI Foundation. Anthropic is donating the Agent Skills standard to the Agentic AI Foundation under the Linux Foundation, the same home it gave MCP. OpenAI gave the foundation its Open Responses spec the same day (our story). Our read: nothing changes for skill authors. Anthropic says: "You can continue to use the standard or contribute to it without joining the AAIF." It adds that "So far, 1 million skills have been published and nearly 280 million installed." Next: "We’ve proposed starting with a roadmap covering versioned releases, a formal proposal process, and security guidance for people who write and install skills." (confirmed)
• Anthropic takes proposals for 8 to 12 week life sciences projects. Anthropic is inviting principal investigators at academic institutions, nonprofits and government labs to propose projects in areas including drug discovery and clinical science. Per the post, guest investigators work directly with Anthropic researchers for 8 to 12 weeks. It says: "Our team will run our models and harnesses at no cost to them." On timing: "Applications will be open for approximately 14 days, and we’ll strive to notify selected participants by email approximately 14 days after applications close." Anthropic asks applicants to put only publicly available information in their materials. (confirmed)
• Claude Code 2.1.309 closes several sandbox gaps. Claude Code 2.1.309 reached npm at 4:34 PM ET Thursday. At 6:11 AM ET Friday, latest and next pointed to 2.1.309 and stable was still on 2.1.308 (earlier versions: our Oct 22 story). The CHANGELOG lists "Fixed sandboxed Bash commands on macOS and Linux running the shell's startup files outside the sandbox before each command" and "Fixed sandboxed commands and auto-accepted edits being able to modify the MCP config file passed with --mcp-config". claude ultrareview now requires workspace trust. Per the CHANGELOG, "it asks on a terminal, else refuses; scripts pass --trust-workspace". Our read: if you rely on the sandbox, update now rather than waiting for stable. (confirmed)

OpenAI#

• Decisions API users: the latest OpenAI SDKs break some type checks. Read the release notes before you bump your SDK. openai-python 3.37.0 (4:03 PM ET Thursday) says: "This minor release includes a breaking change to Decisions types." Question instructions "can now be a string or a list of structured text, image, or JSON parts," so check isinstance(question.instructions, str) first. openai-node 7.43.0 (5:31 PM ET) warns that code such as question.instructions.trim() "can fail when it receives an array," and code that extends the old DecisionInputMessage interface "must use DecisionInputMessage.DecisionInputUserMessage instead." Both say existing string instructions still work. Follows Thursday's Decisions API GA. (confirmed)
• Uber rolls out ChatGPT Work to more than 30,000 employees. Per OpenAI, Uber began a company-wide rollout of ChatGPT Work "to more than 30,000 employees across Uber's global workforce," up from a pilot of 1,000, with GPT-6.1 Sol "deployed as the default capability." Uber's engineers "will also continue using Codex." Sumanth Sukumar, Uber's VP of engineering, told TechCrunch that "some 80%" of pilot users "were non-engineers." Neither company has shared cost or productivity numbers; OpenAI says they "plan to share further insights" later. (confirmed)
• Tibo teases "a small ship" for Friday (unverified). Follow-up to his reset post in our Oct 22 digest. OpenAI's Tibo Sottiaux, whose X bio reads "Codex & ChatGPT @OpenAI," posted Day 18 of his daily updates at 12:46 AM ET Friday, ending "But yes… that's a reset!" A minute later he replied: "Landing tomorrow before 10am PST together with a small ship." He posted at 9:47 PM Pacific Thursday, so our read is before 1 PM ET Friday. He didn't say what lands or what the ship is, and nothing had shipped when we checked at 6:20 AM ET. (staff claim, unverified)
• Lovart's design plugin is now in Codex. Lovart posted at 12:02 PM ET Thursday: "The Lovart Plugin is now live in Codex. Generate brand visuals, ad sets, and videos where you code." @OpenAIDevs reposted it at 3:28 PM ET. Neither post gives plan availability or pricing. (confirmed, partner post)
• WSJ: Signal chats came up before OpenAI fired three safety researchers. The Wall Street Journal reported at 9 PM ET Thursday that OpenAI's security team asked three safety researchers to unlock their personal phones last month and saw "disappearing messages" on Signal chats with an outside AI-safety researcher, a week before the firings. The three have denied leaking, and an OpenAI spokesperson told TechCrunch in early October the firings followed a "pattern of misconduct." (reported)

Known issues on OpenAI's status page, Oct 22 to 23 (times ET, all resolved):

• Oct 22, 2:51 to 4:55 PM: Degraded Performance of Dots on Enterprise. "Users are experiencing failures sending messages to dots," and "the majority of impact is affecting enterprise customers," per the status page.
• Oct 22, 3:36 to 8:07 PM: New ad accounts failing to provision admins. "New ad accounts are failing to assign their creator admin" in Ads Manager and the Ads API, per the status page.
• Oct 22, 11:41 PM to Oct 23, 1:58 AM: Delays in customer support responses. OpenAI pointed customers to email support@openai.com while it worked on the issue, per the status page.

Coming next week#

What to watch from Monday, Oct 26 through Friday, Oct 30. Confirmed means a company post, doc or schedule. Unverified means a hint or an open question with no date.

Confirmed#

• Auto mode by default in most Claude apps. Anthropic's stated timing is next week, with no day given: "Starting next week, sessions across Claude web, desktop, and mobile apps will run in auto mode by default on Pro, Max, and Team plans" (@claudeai). Its article says "most Claude apps." (our story) (Anthropic; confirmed, no day given)
• 1Password sign-in for Grok Bot keeps rolling out. @bot said on Oct 22 the integration "is rolling out to users over the next few days." (our story) (Grok Bot; confirmed, no end date)
• Grok Bot workshops. Cursor Workshops lists "SpaceXSI for Higher Education" (Oct 28), "Grok Bot for Enterprise Teams" (Oct 29), "Grok Bot X Clay" (Oct 29) and "What's New in Grok Bot" (Oct 30). When we checked at 6:41 AM ET Friday, the page showed 5:00 PM UTC for all but the Clay session, which showed 6:30 PM UTC; by our conversion, that's 1:00 PM and 2:30 PM ET. (Grok Bot; confirmed, schedule)
• Later: OpenAI's Nov 20 date. The updated Usage Policies say "Effective: November 20, 2026." See today's column for the route audit. (OpenAI; confirmed)

Unverified#

• A statechart companion skill for pstack. Lauren (@poteto), who works on Grok Bot at SpaceXSI and previously worked at Cursor, per her X bio ("grok @bot at @SpaceXSI," "prev cursor"), posted at 6:33 PM ET Thursday that she's "working on a new companion skill that will help your agents draw these beautiful statecharts/diagrams + check them against the specs you made with /create-formal-verification-skill in pstack!" She wrote "will release soon!" and gave no date, so it isn't usable yet. (Grok Bot staff hint; unverified, not released)
• iMessage for Grok Bot's phone apps. Evan Bacon, Grok's frontend lead, replied at 8:52 PM ET Thursday that the rollout is "split up and landing in two parts. iMessage will require a Mac. Will roll out soon". No date given. (Grok Bot staff hint; unverified)
• Phone apps in release notes or docs. Nothing in the store notes, the changelog or the docs as of 6:29 AM ET Friday. (Grok Bot; watching)
• How a Page refreshes. The changelog says the Bot can "keep up to date" a Page but doesn't say how or how often. (Grok Bot; watching)
• Claude enforcement webhooks. The webhooks page still said "Event delivery isn't live" at 6:28 AM ET Friday. Anthropic gives no date. (Anthropic; not announced)
• Hark usage limits. Adcock said Hark would "re-look" at limits (X). No change or date announced. (Hark; unverified)