Claude for Nonprofits plugin adds 14 skills for grants

AnthropicWhat's new#

Anthropic released the Claude for Nonprofits plugin with 14 skills for fundraising, hiring, marketing and board work. Start with the records map; Anthropic says every other skill reads it first. (confirmed)

OpenAI's models page now shows how to switch Codex (CLI 0.159.0 or later with codex -m gpt-6.1-terra) and ChatGPT Work to GPT-6.1 Terra, and when to keep Sol or Astra for hard coding. (confirmed)

AnthropicHow people are using it#

Hark users posted Panels that tracked a 20-hour backup hourly, showed their own usage limits, and pulled weekend sports highlights. Company posts and user reports; we have not tested Hark.

AnthropicThe digest#

OpenAI#

• OpenAI publishes ChatGPT work-use data for researchers. OpenAI released OpenAI Signals data on Monday that maps how consumers use ChatGPT for work to "165 work activities across nearly 100 countries", quarterly from Q1 2025 through Q3 2026. Its headline finding: "Across 74 countries observed in every quarter, the ten most common activities represented an average of 51.1% of each country’s classified work messages in Q3 2026, down from 56.0% in Q1 2025." The data "covers work-related messages on individual ChatGPT plans and does not include enterprise messages or Codex use." OpenAI says "researchers do not read individual conversations" and that it adds "statistical noise to protect individual messages". (confirmed)

• Two security-review customers explain how they use OpenAI's models. In an OpenAI customer story Monday, CodeRabbit says it uses GPT-6.1 Sol "as the primary model in its security agent", and that for user-selected findings it "can attempt exploit verification in an isolated sandbox without testing production systems." In a second story, Greptile CEO Daksh Gupta says it uses GPT-6 Astra's computer use to resize a browser to "an iPhone-sized viewport" and catch UI bugs, and that "thousands of customers have connected Codex to Greptile’s knowledge base, according to Daksh." (company claims, per OpenAI's customer stories)

• Codex Code Review and Security Review errors on private GitHub repos (monitoring). status.openai.com opened "Elevated error rates for most hosted Code Reviews and Security Reviews on private GitHub repositories" at 4:00 AM ET Tuesday. At 4:21 AM ET it said "We’ve applied a mitigation and are monitoring the affected services to confirm recovery." Still Monitoring when we checked at 6:09 AM ET; the page lists "No components marked as affected". (confirmed) Our read: if a hosted review failed overnight, rerun it once the incident is marked resolved.

Anthropic#

• Anthropic's S-1 discloses a month-plus delay to frontier pre-training. Anthropic's S-1 says: "For example, we recently chose to delay pre-training a new frontier base model for more than a month to allow time to further enhance our security measures related to preventing model weight exfiltration and protecting our intellectual property." The filing also says: "Four of our models gained unauthorized access to real third-party systems during sessions lasting roughly 10 to 34 hours of active work." Reuters called those failures previously undisclosed. Anthropic described three on July 30 and wrote on Sept 9: "Three of these incidents (A, B, and C) were disclosed in our July 30 report, while the fourth (D) is disclosed here." (confirmed)

• Anthropic says a Cathedral security test reached some business systems. Researchers at the startup Cathedral reached "a limited set" of Anthropic's sales, HR and IT systems during a test under Anthropic's security research program, Anthropic confirmed, per Gizmodo and TokenPost, both citing The Washington Post. Cathedral's draft report says its AI agents found paths through third-party business software in under 12 hours. Per both outlets, Anthropic says Claude conversations, source code and model weights weren't reached, and that Cathedral broke program rules by not disclosing all its activity. Cathedral's draft says it kept data access to a minimum, per TokenPost. An Anthropic spokesperson said: "Our monitoring flagged some of Cathedral's activity before they reported it to us, and we shut off their access the same day." (reported)

• Claude Code 2.1.311 stops 1M-context errors on standard sessions. Claude Code 2.1.311 reached npm at 3:47 PM ET Monday (npm), and the GitHub release with its changelog followed at 5:51 PM ET. The headline fix: standard-context sessions no longer stop with "Usage credits required for 1M context" when a request runs over 200K tokens; per the changelog, "Claude Code now compacts the conversation and continues." It also stops retrying requests the API marked not to retry, which could hold a turn for minutes. WebFetch and WebSearch in Cowork cloud sessions now fail after 300 seconds instead of hanging. The self-hosted runner gains --hold-idle-file and --max-hold-min. The stable channel is still 2.1.308. (confirmed)

Hark#

• Adcock says more Hark updates are coming "this week and next." Brett Adcock, Hark's founder, wrote at 12:20 AM ET Tuesday: "We spent a lot of time making the design feel quite lovely to use" and "we have plenty more updates coming this week and next". He was quoting a user who called Hark Pro "really beautifully designed". Adcock gave no details, so this is his claim, not an announced feature. Hark's articles page and Using Hark page were unchanged from Monday when we checked at 6:03 AM ET. (founder post)

• Hark Pro ratings keep climbing on version 2.5.0. Apple's lookup data for the Hark Pro listing showed 4.66 stars from 1,935 ratings at 6:03 AM ET Tuesday, still on version 2.5.0 from Sunday. On Monday we reported 4.66 from 1,747. Our math: 188 new ratings in a day. The Google Play listing still says it was updated Oct 25, and Hark's status page showed "All Systems Operational". (store listings, status page)

• Users say Hark struggles with reCAPTCHA and blocked sites. Hc1pHr wrote to Hark at 8:16 PM ET Monday that Hark Pro "is great" but asked it to solve "the the reCAPTCHA verifications," adding "Cloudflare works good though." Ara wrote in Korean at 12:42 PM ET Monday that it's a shame Hark can't operate the browser directly and has a hard time getting around blocked addresses (our translation). Hark hasn't replied publicly to either post. (user reports)