Google announces Gemini 4 Carbon, initially only for security teams
Google calls Gemini 4 Carbon its most capable model, with an emphasis on cybersecurity and long, multi-step tasks. According to Google, the model scores 88.2% on SEC-bench Pro (finding vulnerabilities) and 71.0% on CWE-bench (closing leaks). This week only a portion of cyber defenders gain access; developers, businesses and consumers will follow "as soon as possible", with no date given. The introductory API price is $2 per million input tokens and $10 per million output tokens, after which it rises to $4 and $20.
What does this mean? You cannot yet use Gemini 4 Carbon yourself; factor in the stated API prices if you are considering a switch, as they double after the introductory period.
Claude for Legal gets a gallery of ready-made legal artefacts
Anthropic has added a gallery of interactive artefacts to Legora for Legal, including a cite check, tabular review, matter board and a scan for upcoming regulation. In addition, there are artefacts from Harvey, PwC, Goodwin and Foley & Lardner; according to Anthropic, Claude and Orrick will follow. Each artefact starts with an example, after which you connect your own documents and connectors; since May, more than 60 legal connectors have been added.
What does this mean? If you work with Claude for Legal, you can start contract review and source checking from a fixed template instead of a blank chat; the legal review remains your responsibility.
ChatGPT can now read and update iCloud Mail and Apple Calendar
OpenAI has added plugins for iCloud Mail and iCloud Calendar to ChatGPT on the web, the desktop app and iOS, across all subscriptions. With them, ChatGPT can search and read mail, manage drafts, send email, look up contacts and schedule or change appointments. Connecting works via an app-specific password and only applies to addresses at @icloud.com, @me.com and @mac.com.
What does this mean? If you connect a mailbox containing client or patient data, you give ChatGPT access to personal data; only do so if it fits within your own privacy arrangements.
38 municipal websites unintentionally unreachable for AI assistants
Research by AI-Checker.nl across all 342 municipalities found that on 13 October, 38 municipal websites were not reachable for certain AI crawlers, while ordinary visitors and Google could still get in. For 31 municipalities, this was caused by default settings in Microsoft Azure Front Door, which blocked unknown bots. On 26 October, seven municipalities were still blocked, so ChatGPT may give outdated information about those municipalities.
What does this mean? If you want AI assistants to use up-to-date information from your own site, check that your hosting or security service is not quietly blocking AI crawlers.
