September 21–27 brought new ways to work by voice, browser updates and a revealing example of AI-assisted fraud. This issue covers ten developments, with primary sources and a short explanation of their practical relevance. Dates below refer to announcements or releases; availability conditions apply to the specific feature described.
1. Claude Opus 5.5: a new model and lower API costs
On September 22, Anthropic introduced Opus 5.5. The company claims performance close to Fable 5.1 on most tasks and costs 40% below Opus 5 on typical workloads at default settings. That is the vendor’s estimate, not a promise of identical savings in every case.
Compare the cost of a completed task, including review and retries. Our separate Opus 5.5 explainer covers API rates and why they are different from a chat subscription price.
2. Googlebook pre-orders are open
On September 21, Google opened Googlebook pre-orders for laptops combining Android with ChromeOS desktop foundations. The US starting price is $899. Sales are scheduled for October 4 in the US and October 5 in the six other countries listed in the announcement.
Before buying, check your essential apps, phone compatibility and local sale conditions. Our Googlebook article explains the Android requirements and AI feature restrictions.
3. ChatGPT brings plugins into Live
On September 23, OpenAI added plugins to Live on web, iOS and Android. Voice conversations can use the connected apps available to your account. Existing permissions and usage limits still apply.
A useful starting point is a short request about one document, followed by checking its original link. Our ChatGPT Voice plugins article explains connections, action approval and availability.
4. Gemini 3.8 TTS adds more control over generated speech
On September 23, Google announced Flash TTS and Flash-Lite TTS for generating speech from text. Rollout started in the Gemini API and AI Studio, with Enterprise API access coming later. Flash TTS is also reaching Gemini Notebook, while Flash-Lite TTS is coming to Google Vids. Google claims support for more than 100 languages.
For a voiceover, first check a short passage in your target language: names, stress and numbers. A language appearing in a support list does not establish the quality of your particular script.
5. Live Avatar adds a visual conversational agent for businesses
On September 24, Google introduced Gemini 3.8 Live with Live Avatar in Gemini Enterprise. It combines speech with a generated video character; the company claims speech synchronization across 97 languages. Creating custom avatars requires separate enterprise allowlisting.
This is an enterprise offering, so the announcement alone does not establish that the same control exists in personal Gemini accounts. Useful evaluation criteria include answer accuracy and how easily a question can be handed to a person.
6. Chrome 154 arrives with security fixes
On September 22, Google started rolling out Chrome 154 in the stable channel for Windows, macOS and Linux. The notes list 108 security fixes, including critical issues. Rollout is gradual.
Check the update offered in your stable channel and complete a restart if requested. A higher version number alone is not a reason to switch to a testing build.
7. Firefox 156.0.1 fixes hangs and interface problems
On September 22, Mozilla released Firefox 156.0.1. Fixes address hangs on some pages using CSS anchor positioning, NVDA behavior around address-bar buttons, and window behavior on macOS 27. Tab now goes straight to the text field in the address or search bar.
This release is particularly relevant if familiar keyboard navigation changed or certain pages became unresponsive after a previous update. The description of a fixed problem is more useful than a generic claim that a browser is faster.
8. Microsoft reports disrupting EvilTokens
On September 22, Microsoft described its operation against EvilTokens. The company linked the service to more than 12,000 compromised inboxes and said it used AI to analyze email and prepare fraud. Victims were persuaded to enter a code on Microsoft’s genuine sign-in page, effectively granting outsiders access to their accounts.
A genuine sign-in address does not, by itself, explain who receives access. Do not enter a connection code you did not initiate; verify an unusual request in a work email through another known contact channel.
9. ChatGPT gains a security event history
On September 25, Security history was announced in ChatGPT.
Security history shows sign-ins, sign-outs and changes to protective settings. On the web, open Settings → Security and login → Security history. Location and device details may be approximate or missing; active sessions are managed separately.
Review the event time and type alongside the device, instead of drawing a conclusion from a city name alone. If you see an action you did not perform, follow the account-security instructions in the help article.
10. Project Suncatcher prepares to test AI chips in orbit
On September 24, Google outlined an upcoming Project Suncatcher test: a prototype satellite will evaluate TPUs in space. The mission is being prepared with Planet for Transporter-18; the team is studying radiation, vibration and heat removal. A communications test between two satellites is planned for 2027.
This is a research stage with questions still to answer. The announcement does not establish that an industrial orbital data center is operating or that the economics of such infrastructure have been proven.
What to do this week
- Check browser updates and review unfamiliar activity in important accounts.
- If voice apps are available to you, try one small task and compare the result with its source.
- Distinguish available features from pre-orders, enterprise approvals and future experiments.
Previous issue: the September 14–20 digest.
Join the conversation
Stay on topic and respect other readers. Your first comment may appear after editorial review.