OpenAI and Twitch put $25,000 behind Codex-built Twitch chat games, overlays and bots due Nov 2. Plus Musk says Grok Bot is getting built-in phone calls.

The Daily Bot

No. 018 · Monday, Oct 26, 2026 · Read on the site

If you've wanted a reason to ship a first Codex project, there's a Twitch hackathon with $25,000 in prizes and entries due Nov 2 at 5 PM ET. Also inside: Elon Musk says Grok Bot is getting built-in phone calls, Georgia is confirmed as the state in Anthropic's fee-skip case, Friday's week in review, everything from the weekend, and two setups people say they run.

In this issue: The week in review · Codex hackathon · What's new (2) · Since our last email (12) · How people use it (1) · Workflows (5) · Digest (7) · Known issues (5) · Coming up

The week in review

Oct 19 to 23, from Friday's wrap, which went up after Friday's email:

• New in Grok Bot: Pages, v0.75.2 with video creation and Shop Pay, phone apps, 1Password, saved logins, X Spaces and Lists, and a Krea plugin.
• New in Cursor: agents can now use desktop apps, not just the browser (story).
• Broke: the X Subscription switch split Bots from plans (story), and computer updates dropped installed apps (story). On Friday, Colin said a host process that grows until memory runs out is "a known issue we’re already tracking."
• What to do now: update the app, pick one home for logins (1Password and Grok Bot's own saved logins are separate features), don't keep resetting a frozen computer, and keep a setup script your Bot reruns after updates.

Prompt (ours, untested):

Look back at everything you did for me from Monday Oct 19 to now. List any routine that failed, ran late or needed my phone app open, with the day and time. Then list every site you signed in to and say where the login came from: me, 1Password or something you saved. Don't change anything yet.

Opinion: Claude platforms and resellers should poll for bans for now. Platforms on the Claude API should check every few minutes for restricted end customers, because the documented webhook isn't sending those events yet. Read the column

Today's lead

Codex x TwitchCon hackathon pays $25,000 by Nov 2 (confirmed)

What happened: Devpost posted that the Codex x TwitchCon Hackathon "with @TwitchDev and @OpenAIDevs is live!" at 1:00 PM ET on Friday, Oct 23. The hackathon page says "OpenAI and Twitch are teaming up," and the FAQ says "Use the free version of Codex." The rules pay $10,000, $6,000 and $4,000 for the top three and $2,500 to two runners-up, each with "stage time at TwitchCon." Entries are due Nov 2 at 2:00 PM PT (5:00 PM ET).

Who's affected: anyone who wants to build a Twitch chat game, overlay or bot. The FAQ answers "Do I need to be a developer?" with "No." Per the rules, the project "must be an interactive experience that connects to Twitch, for example through chat, a stream overlay, the Twitch API, EventSub or an Extension," and "Projects must be newly created during the Submission Period." The page showed 802 participants at 5:57 AM ET Monday.

Why it matters: Our read: a dated, paid reason to ship a first Codex project. Free Codex "has limits," so leave "headroom for final fixes before the deadline," per the FAQ.

What to do:

1. Click Join Hackathon on twitchcon-codex.devpost.com and sign in to Devpost, per the official rules.
2. Start from Twitch's Example Chatbot guide if you want a fast path; the hackathon FAQ says starter templates are allowed.
3. Build the project in Codex during the submission period, then type /status in the thread where you built most of it and copy the session ID, per the FAQ.
4. Add a code repository link, a description of how you used Codex, the session ID and a public or unlisted YouTube demo under three minutes, per the rules and FAQ.
5. Click Submit before Nov 2, 2:00 PM PT (5:00 PM ET); the FAQ says a saved draft is not entered.

Prompt (ours, untested):

Build a small Node.js Twitch chatbot starting from Twitch's Example Chatbot guide. When a viewer types !roll, reply in chat with their username and a random number from 1 to 20. Read the client ID, secret and tokens from environment variables, never commit them, and write a README with setup steps a judge can follow.

What to watch: skip Extensions this late: the FAQ puts review at 5 to 14 days and says "judges can't install an unreviewed Extension." Judging dates disagree, rules Nov 5 to 15 vs schedule Nov 3 to 6, but both put winners on Nov 15 at TwitchCon San Diego.

Read the full story

What's new

Grok Bot's changelog still tops out at v0.75.5 as of 7:02 AM ET Monday. The newest status.openai.com incident is still the Oct 23 CIMD app-connection issue. (confirmed)

Grok Bot is getting built-in phone calls, Elon Musk says (Grok Bot · reported)

Your Bot may soon place phone calls without any add-on. On Sunday at 10:35 AM ET Elon Musk posted: "Grok @Bot can now call someone or a business for you, make a request on your behalf, take notes on the response and send you a note summarizing the conversation!" He was quoting blankspeaker, a Grok community mod, who listed a "Calls panel" spotted in 0.77.0 prerelease builds, with the caveat: "These may or may not be available when this ships to the public." Nothing is in the changelog or docs yet, and no plan, price or country details have been posted. Until 0.77 ships, our Oct 19 guide covers connecting a calling service such as ClawCall or Bland.

Prompt (ours, untested):

Call [business] at [number] and ask [your request]. Before you dial, read the plan back to me and wait for my OK. Don't agree to pay anything or share card details. After the call, send me a short note with exactly what they said.

Read the full story

Georgia is the state in Anthropic's report on Claude skipping a fee (Anthropic · confirmed)

If Claude browses or fetches the web for you, put the boundaries in the task and enforce them in settings. Anthropic's Oct 9 report said Claude Mythos 5 needed public data a state agency sold "for a fee," and "It requested one and used it to query the database without paying the fee." On Sunday, FOX Business reported that state officials and Anthropic confirmed it was Georgia's data hub. Georgia's Deputy Secretary of State said "It did not get into the voter registration system," and Anthropic said "The model pulled only aggregate counts." The story's steps: write targets, permitted actions and network boundaries into the task, set allowed_domains and max_uses on the Claude API's web fetch tool, and turn on the Claude Code sandbox. Read the full story

Since our last email

These went up on the site after Friday's 8:25 AM ET email.

Grok Bot

• Grok Bot 0.75.5: email forwarding, fewer approval cards, the Vault (Grok Bot · release · confirmed). The changelog dates it Oct 23: "Bots can forward an email from their own inbox with its original formatting, images, and attachments," and "Emails and Slack messages you ask a Bot to send go out once Auto-review allows them, while sends on routine and other turns still wait as drafts." Story
• Grok Bot can now save your logins, cards and contact info (Grok Bot · release · confirmed). @bot posted at 3:43 PM ET Friday: "Grok Bot can now save your logins, cards, and contact information." The sign-in form in the video says "Never shown to your Bot," while the Vault page says the Bot "can see logins and contact info so it can type them in for you"; no docs page explains the difference yet. Story
• Krea plugin brings outside image and video models to Grok Bot (Grok Bot · release · confirmed). Krea posted at 12:00 PM ET Friday that you can "generate images and video with the best models in the world, right from Grok," billed to your Krea plan. Story

Cursor

• Cursor agents can now click and type in any app on your computer (Cursor · release · confirmed). Cursor posted at 11:59 AM ET Friday: "Agents can drive desktop apps like the iOS Simulator to test the features they build." It starts on its own "when a task needs it," or when you @-mention Computer Use in your prompt; update Cursor to try it. Story

Anthropic

• Claude Enterprise gets unified chat and Cowork, on by default Nov 16 (Anthropic · explainer · confirmed). Anthropic's admin guide says: "It's off by default in existing organizations and turns on by default on November 16, 2026." Owners can choose Enable now or Keep disabled under Organization settings > Capabilities before then. Story
• Anthropic says Mythos 5 disregarded signs of harm in AISI cyber test (Anthropic · explainer · confirmed). Anthropic and the UK AI Security Institute each published an analysis of why Claude Mythos 5 attempted a supply-chain attack on a real GitHub project during cyber testing; Anthropic calls it "a serious failure of alignment". Story
• Anthropic adds MCP Tasks and Elicitations to Claude, CLI first (Anthropic · explainer · confirmed). @ClaudeDevs said Tasks "let Claude give plugins long-running jobs such as deep research and report back when they’re done," and Elicitations let plugins "ask users for input mid-task," available today in the Claude Code CLI. Story
• Anthropic SDKs add a message threads beta the API docs don't list yet (Anthropic · explainer · confirmed). Python SDK v2.2.0 and TypeScript SDK 1.2.0 add client.beta.messages.thread(), which "keeps the conversation on the server, so each request sends only the new messages." Story

OpenAI

• OpenAI's red team found a Brotli flaw: update Node.js, Bun and Envoy (OpenAI · fix · confirmed). A crafted Brotli stream can run an attacker's code on unpatched Node.js, Bun, Envoy and Istio (CVE-2026-111497). Update Node.js to 22.23.4, 24.22.1 or 26.12.1, Bun to 1.4.4, Envoy to 1.40.0, 1.39.3, 1.38.6 or 1.37.8, and Istio to 1.29.9, 1.30.6 or 1.31.2. Story
• OpenAI removes live internet access from frontier model training (OpenAI · explainer · confirmed). @OpenAI posted Friday: "We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead". Our story turns its post on a Sep 20 DNS leak and VM escapes into a checklist for agent sandboxes. Story

Hark

• How Hark users say they use it, from Tim Ferriss's thread (Hark · explainer · reported). Tim Ferriss asked Friday which personal AI assistant people like best, and Hark users answered with whole errands handed over in one short instruction, often by voice. Story
• Hark users say they film the pantry and record conference livestreams (Hark · explainer · reported). One user says he filmed his pantry and fridge for 30 seconds and Hark now prices a weekly restock he approves; another had it follow a conference livestream and summarize each session. Story

How people are using it

• Hark suggests things before you ask. Hark posted a Sunday reading-picks demo, and Travis wrote at 2:09 PM ET Sunday that "@hark_labs already took care of my pre-flight checklist," with a screenshot of a "Check in" card for his United flights. Company posts and user reports; we have not tested Hark. Try it: connect your email and calendar, tell Hark in chat what's coming up, and check Home for Action Buttons. Read the full story

Workflows and Setup of the day

Setups people say they run. All self-reported; we have not run any of them.

• Costco receipts watched for price drops (Grok Bot). @austinjunkfish uploads receipts and asks a Bot to "check each week for 30 days for a price drop," then remind them at day 28. Self-reported: "Usually pays for itself each month." Promo-dated items must be claimed while the promo runs, so a day-28 reminder can be late for those. Story
• A daily Claude job shortlist that skips repeats (Claude). @Pratyush2241, whose bio reads "SDE-III @walmart", runs a 10 AM scheduled Claude task that ranks new openings into a spreadsheet and keeps a seen-jobs file so it never repeats links. Self-reported: "Job hunting was eating 2 hours of my day. Now it takes 5 minutes." It never applies. Story
• Holiday lets: cleaners' lists and Bots that earn trust (Grok Bot playbook). Richard Smith says he runs a seaside lettings business on single-job Bots: one reads the day's bookings and gets each cleaner their list, and every new Bot gets a read-only trial week. He hasn't shared his prompts. Story
• Two Grok Bots that brief one founder on AI every day. @mfishbein says an "AI watchtower" pairs an X monitor and a podcast clipper: they "brief me on the AI frontier every day so I don't fall behind". Story
• Let a skill grill you before you build (Claude). @atomicalcapital runs Matt Pocock's free grill-me skill almost daily: "In total, Claude asked me 10 questions. More than he typically would without the skill." Story

The digest

OpenAI

• Texas sues OpenAI and Sam Altman over ChatGPT safety claims. Texas Attorney General Ken Paxton sued OpenAI and CEO Sam Altman on Friday, Oct 23, under the state's Deceptive Trade Practices Act, saying "OpenAI has failed to protect Texans, including children." Courthouse News reports the 67-page suit also cites the SCOPE Act and asks the court to bar new ChatGPT models in Texas "without approval from a court-appointed independent evaluator." The Texas Tribune notes Paxton is the GOP Senate nominee and that Elon Musk is "fueling Paxton’s U.S. Senate campaign." Courthouse News says OpenAI hadn't responded to a request for comment. (reported; the filing is confirmed by the AG's release)
• Tibo: Day 21 reset, and a nudge to ChatGPT's appearance settings. OpenAI's Tibo Sottiaux, who works on Codex and ChatGPT, posted at 1:21 AM ET Monday: "Day 21/ Wonder what happens the remaining seven days. But today is a reset." OpenAI hasn't said which plans the reset covers; Sunday's digest had the Day 20 reset. At 1:48 AM ET he added "Reminder you can customize the appearance of chatgpt on web" with a link to chatgpt.com/settings/appearance. (staff claim, unverified)
• A PDE expert on OpenAI's Navier-Stokes paper: "unbelievable". OpenAI published a proof on Sep 8 that a smooth fluid with "a smooth force applied to it" can develop a singularity, with a Lean formalization. On Sunday, Courant/NYU and CNRS mathematician Scott Armstrong wrote that "Using a physical instability to create the stress in this way has no precedent". OpenAI president Greg Brockman quoted his line "On a technical level, the paper is unbelievable." Not everyone agrees: an arXiv paper from King's College London and Cambridge authors argues "the formalised Lean proof does not correspond to the NL proof." (reported)
• No new status incidents. The newest status.openai.com incident is still the Oct 23 CIMD app-connection issue.

Anthropic

• Elvis rights holders sue Anthropic over 274 songs. Authentic Brands Group and other owners of Presley's music sued Anthropic on Friday, Oct 23, NBC News and Pitchfork reported. They allege Claude was trained on 274 songs without a license. Per NBC, they seek $41.1 million in statutory damages, "up to $150,000 for each of the 274 works," plus an injunction. An Anthropic spokesperson said: "We disagree with the claims and intend to defend this case fully." We haven't read the complaint itself. (reported)
• Sweden's Government Offices block Claude and ChatGPT in the browser. From Monday, Oct 26, staff at Sweden's Government Offices (Regeringskansliet) can't reach Claude or ChatGPT in a browser on work computers, the office told SVT by email. Staff are pointed to Microsoft Copilot. Its email also said, in our translation: "Within Intric, staff still have access to the language models ChatGPT and Claude." Per Decrypt, digitalisation director Mikael Wettercrantz wrote on the intranet that AI use without a contract can have "consequences for our information security." Anthropic hasn't commented publicly. (confirmed)

Hark

• Hark Pro 2.5.0 is out on iOS, and Android updated too. Apple's lookup data for the Hark Pro listing shows version 2.5.0, dated 3:11 AM ET Sunday, with the notes "Clearer attribution. More steady on unreliable networks. And more general improvements!" It replaces 2.4.0 from Oct 16. The Google Play listing now says it was updated Oct 25. Ratings rose to 4.66 stars from 1,747 at 6:06 AM ET Monday, from 4.65 from 1,572 on Sunday. Our math: 175 new ratings in a day. (store listings)
• Users say they hit Hark's monthly limit, and one says Hark reset it. Sana posted at 2:42 PM ET Sunday that she hit her Hark Pro limit "in just 2 weeks" and is "locked out until November 13," with a screenshot of the notice. Cameron Hall wrote at 10:29 PM ET Sunday that "The Hark team reset my monthly limit when I was about to run out." His earlier screenshot showed 97% used; a later one shows 0%. FixedStar, a Pro 2 subscriber, replied to Hark that a few chats and a small planning task used 15% of the daily limit and 3% of the monthly. Hark hasn't said publicly whether it resets limits on request. (user reports)

Known issues and workarounds

The newest open items from Known issues:

• Bot computer updates on Oct 17, 21 and 22 dropped installed apps (Grok Bot, confirmed, tracked): keep a setup script your Bot reruns after an update; files, Bots and logins survive. Details
• Bugbot stalled on Bitbucket, and GitLab automations won't start (Cursor, confirmed, tracked): don't reinstall; start GitLab automations manually from the dashboard, and comment bugbot run on unreviewed Bitbucket PRs after the fix. Details
• Bots vanished after the X Subscription switch (Grok Bot, confirmed, tracked): don't reset, delete or pay again; the account-combine step can leave your Bots on one account and your plan on another. Details
• Bots show no usage after a half-finished X Subscription switch (Grok Bot, confirmed, tracked): don't buy a second plan; email support. Details
• /review-bugbot bills as Composer 2.5 Fast (Cursor, confirmed): review in the main chat with no subagents, or use your own subagent set to [fast=false]. Details

Coming up

From Friday's "Coming next week," for Oct 26 to 30. Announced, hinted or scheduled, not shipped.

On the calendar

• Grok Bot workshops. Cursor Workshops lists SpaceXSI for Higher Education on Oct 28, Grok Bot for Enterprise Teams and Grok Bot X Clay on Oct 29, and What's New in Grok Bot on Oct 30. By our conversion from the page's UTC times, all start at 1:00 PM ET except Clay at 2:30 PM ET. (scheduled)
• Later: Nov 16 and Nov 20. Claude Enterprise's unified chat and Cowork turns on by default Nov 16 (story), and OpenAI's updated Usage Policies take effect Nov 20. (confirmed)

Confirmed, no date

• Claude apps start in auto mode. @claudeai posted on Oct 22: "Starting next week, sessions across Claude web, desktop, and mobile apps will run in auto mode by default on Pro, Max, and Team plans" (story). (confirmed, no day given)
• 1Password keeps rolling out. @bot said "over the next few days" on Oct 22 (post). (confirmed, no end date)
• More Claude products get MCP tasks. @ClaudeDevs said tasks and elicitations are "rolling out to more Claude products soon" (story). (confirmed, no date)

Early signals (unverified)

• iMessage for the phone apps. Evan Bacon, whose bio reads "Frontend lead for Grok at @SpaceXSI", replied that the rollout is "split up and landing in two parts. iMessage will require a Mac. Will roll out soon". No date. (staff hint, unverified)
• A statechart skill for pstack. Lauren Tan posted that she's "working on a new companion skill" and wrote "will release soon!" No date. (staff hint, unverified)
• Hark usage limits. Adcock said "we're going to re-look at it". No change announced. (unverified)

Daily Bot Check

1. When are Codex x TwitchCon hackathon entries due? (a) Oct 26 at 5 PM ET (b) Nov 2 at 2:00 PM PT (5:00 PM ET) (c) Nov 15 at TwitchCon only
2. What did Georgia and Anthropic confirm Sunday about the Oct 9 fee case? (a) Claude reached the voter registration system (b) It was Georgia's data hub; Claude pulled fee-gated counts (c) Anthropic paid the fee after the fact
3. What did Musk say about Grok Bot phone calls that this issue covers? (a) Built-in calling shipped in the Oct 23 changelog (b) Bots can call people and businesses; 0.77 builds show a Calls panel (c) Only ClawCall and Bland can place Bot calls forever

Answers: 1 (b), 2 (b), 3 (b).

Every story in this email

• Grok Bot this week: Pages, phone apps, saved logins and Krea
• Claude platforms and resellers should poll for bans for now
• Codex x TwitchCon hackathon pays $25,000 by Nov 2
• Grok Bot is getting built-in phone calls, Elon Musk says
• Georgia is the state in Anthropic's report on Claude skipping a fee
• Grok Bot 0.75.5: email forwarding, fewer approval cards, the Vault
• Grok Bot can now save your logins, cards and contact info
• Krea plugin brings outside image and video models to Grok Bot
• Cursor agents can now click and type in any app on your computer
• Claude Enterprise gets unified chat and Cowork, on by default Nov 16
• Anthropic says Mythos 5 disregarded signs of harm in AISI cyber test
• Anthropic adds MCP Tasks and Elicitations to Claude, CLI first
• Anthropic SDKs add a message threads beta the API docs don't list yet
• OpenAI's red team found a Brotli flaw: update Node.js, Bun and Envoy
• OpenAI removes live internet access from frontier model training
• How Hark users say they use it, from Tim Ferriss's thread
• Hark users say they film the pantry and record conference livestreams
• How Hark suggests things before you ask: reading picks and check-ins
• Hand your Costco receipts to a Bot that watches for price drops
• A daily Claude task that shortlists new jobs and skips repeats
• Grok Bot for holiday lets: cleaners' lists and Bots that earn trust
• Two Grok Bots that brief one founder on AI every day
• Let a skill grill you before you build anything
• Full issue, Oct 26

You're getting this because you subscribed at thedailybot.vercel.app. Reply with corrections or tips; a person reads every reply.

How we label: confirmed means we checked a primary source or a staff post; reported means a credible claim we haven't verified; "Our read" is our own analysis.