23 October 2026
Claude connects to Google Docs, Sheets and Slides as agent security tools arrive
Today's news is about AI tools reaching deeper into everyday work apps. Security is a big theme, with new agent login controls, a Meta sharing leak and a serious web bug. There is also news on image models and OpenAI's funding.
Security and safety
1Password lets AI agents log in without seeing passwords (opens the source in a new tab)
1Password has launched a feature that lets agents log into a customer's personal accounts. The agent does not see the private credentials.
Why it matters: If you plan to let agents use your accounts, never paste passwords into prompts and look at tools like this first.
TechCrunch · 22 Oct
Meta turns Muse sharing back on after privacy report (opens the source in a new tab)
Meta has re-enabled Muse sharing. Researchers at Claroc had reported that public conversation links were easy to find, including a business revenue review and a loan application.
Why it matters: Check what your team shares through AI chat links, because a public link can be found by strangers.
The Verge · 22 Oct
Hackers got a four-day head start on PHP2Shell bug (opens the source in a new tab)
Hackers were probing the PHP2Shell bug before defenders were told it existed.
Why it matters: If your website runs on PHP, ask your developer or host whether they have patched this bug.
Ars Technica · 22 Oct
Legal experts say Meta's AI could still scrape news sites (opens the source in a new tab)
Meta changed its business terms to remove language about training AI on news sites. Legal experts say the change does not go far enough.
Why it matters: If you publish content online, check the terms of any platform where you post.
The Verge · 22 Oct
Tools and products
Claude now works with Google Docs, Sheets and Slides (opens the source in a new tab)
Claude can now work with Google Docs, Sheets and Slides. The title gives no further detail.
Why it matters: If your team lives in Google Workspace, check whether Claude can now handle drafts and reports without copying and pasting.
claude.com via Hacker News · 22 Oct
OpenAI launches the Decisions API (opens the source in a new tab)
OpenAI describes the Decisions API as fast, focused, low-cost decisions for any application.
Why it matters: If you build apps on OpenAI, look at the pricing for simple decision steps before your next invoice.
OpenAI · 22 Oct
Canva's Craft image model plans designs before generating (opens the source in a new tab)
Canva Craft uses two stages, so it plans a design before it makes the image. Canva says this gives greater precision, consistency and control.
Why it matters: If you make client graphics in Canva, test whether the planning step keeps your layouts on brand.
The Verge · 22 Oct
Agents
OpenShell: a safe, private runtime for AI agent fleets (opens the source in a new tab)
OpenShell is a project on GitHub described as a safe, private runtime for fleets of autonomous AI agents. It drew high interest on Hacker News.
Why it matters: If you plan to run several agents, ask your developer how they will be kept safe and private.
github.com via Hacker News · 22 Oct
n8n compares stateful and stateless AI agent design (opens the source in a new tab)
n8n explains how stateful and stateless design change scaling and memory of context. It covers memory nodes and state stores.
Why it matters: Decide early whether your agent needs to remember past chats, because it changes how you build it.
n8n · 22 Oct
Meta pushes a future where AI agents talk to each other (opens the source in a new tab)
The Verge describes Mark Zuckerberg's vision of autonomous AI agents that shop, work and talk to other people's agents.
Why it matters: Customers may soon deal with your business through their own bots, so keep your information clear and easy to read.
The Verge · 22 Oct
Automation
n8n explains how to detect hallucinations in agent workflows (opens the source in a new tab)
n8n says agent hallucinations are hard to catch because they look like normal answers. A mistake can then trigger other workflow steps and become a production problem.
Why it matters: Add a check step before an agent's output triggers anything that reaches customers.
n8n · 22 Oct
n8n guide: route between self-hosted LLMs and cloud APIs (opens the source in a new tab)
n8n says a self-hosted LLM is not always cheaper than an API. Its guide covers when running models yourself wins and how to route between local and cloud models.
Why it matters: Before buying hardware to run models yourself, compare the real cost with API pricing.
n8n · 22 Oct
n8n guide to API versioning for stable integrations (opens the source in a new tab)
n8n compares versioning methods such as URI, header and date-based. It also covers how to handle a provider ending support for a version.
Why it matters: Check which API versions your automations use so a provider change does not break them.
n8n · 22 Oct
Enterprise
OpenAI reportedly working on $30B raise at $1.25T valuation (opens the source in a new tab)
TechCrunch reports OpenAI is working on a new raise ahead of an IPO. It would make OpenAI the most highly valued private company ever.
Why it matters: Heavy funding suggests OpenAI will keep building fast, but avoid relying on a single vendor.
TechCrunch · 22 Oct
OpenAI and Poland's NASK partner on AI safety (opens the source in a new tab)
OpenAI and Poland's NASK are partnering on AI safety, security and technical research. They aim to strengthen safeguards.
Why it matters: Safety work between governments and AI vendors can shape future rules, so watch it if you serve regulated clients.
OpenAI · 23 Oct
Enterprise AI advantage lies in what you build around the model (opens the source in a new tab)
MIT Technology Review argues that the next advantage in enterprise AI has less to do with which model you choose than with what you build around it.
Why it matters: Spend time on your data, checks and workflows, not only on picking the newest model.
Research
Anthropic says LLMs make formal software verification feasible at larger scale (opens the source in a new tab)
A post on anthropic.com says large language models are making formal software verification feasible at a much larger scale.
Why it matters: Over time this could mean fewer bugs in software you rely on, but it is not something to act on today.
anthropic.com via Hacker News · 22 Oct
Same AI request emits 8.5 times more CO2 in Texas than Sweden (opens the source in a new tab)
A berget.ai post says the same AI request produces 8.5 times more CO2 in Texas than in Sweden.
Why it matters: If clients ask about your carbon footprint, the region where your AI runs may matter.
berget.ai via Hacker News · 23 Oct
OlmoEarth v1.3 offers embeddings for the whole planet (opens the source in a new tab)
Hugging Face has published OlmoEarth v1.3, described as state-of-the-art embeddings for the whole planet.
Why it matters: This mostly matters if you work with maps or land data, otherwise you can skip it.
Hugging Face · 22 Oct
Also worth knowing
Most Americans say AI is bad for society (opens the source in a new tab)
ABC News reports most Americans say AI is bad for society and oppose more data centers.
Why it matters: Customer attitudes to AI are mixed, so be open about where you use it.
abcnews.com via Hacker News · 22 Oct