Last checked November 17, 2026 · News roundup
Today in brief: Google starts switching Gemini Gems to skills today, and personal-account Gems turn read-only before they disappear on December 17. Microsoft Ignite opens in San Francisco with a new Copilot Plugin Catalog, while OpenAI added 24-hour ChatGPT suspensions for child-exploitation prompts and the FBI warned about Artex, an AI hacking tool. On the business side: what broke at Chase on Monday, near-sevenfold demand for Anthropic's IPO, and $178 million for two AI startups.
Today's eight stories at a glance
| Story | Date | Who it hits |
|---|---|---|
| Gemini Gems become skills | Nov 17 (removal Dec 17) | Personal Google accounts, 18+ |
| Microsoft Ignite + Copilot Plugin Catalog | Nov 17–20 | Microsoft 365 Copilot users and admins |
| ChatGPT child-safety suspensions | Nov 16 | ChatGPT users, API developers |
| FBI warning on Artex | Nov 16 | IT and security teams |
| Apollo Research inside OpenAI | Nov 16 | Enterprise AI buyers, policy watchers |
| Chase online banking outage | Nov 16 | Chase customers, anyone sending them money |
| Anthropic IPO demand | Pricing Nov 18 | Investors, Claude business customers |
| Corma and Dreamteam funding | Nov 16 | Security and sales teams shopping for AI tools |
AI
Gemini Gems are now skills: what changed today and what to save by December 17
Starts Nov 17 · personal Google accounts, 18+ · no price change announced
As of November 17, Gems in the Gemini app become read-only for adults on personal Google accounts. You can still open a Gem to look at it, but you can't chat with it, edit it or make a new one. Google converts each Gem into a skill automatically, and the old Gems are deleted on December 17, 2026, according to Google's help page on the transition.
Skills work differently. Instead of opening a Gem from the sidebar, you type a slash and the skill name in any chat, or let Gemini apply a skill on its own when it fits your prompt. You can stack several skills in one chat, keep up to 100 active at once, and on the web app or Mac app you can upload SKILL.md files built for other platforms. If you plan to import third-party skills, read our guide on checking a skill file before you install it; the same format carries the same risks.
Two things do not carry over cleanly. Skills hold 100 MB of files in total, and some file types are dropped. Most of the tools a Gem could default to, including Canvas and Deep Research, don't work with skills.
When your Gems move, and what survives the move
Nov 17, 2026
Personal accounts (18+): Gems turn read-only, skills take over
Dec 17, 2026
Personal-account Gems deleted; download files before this
Mar 2027
Workspace business, enterprise and nonprofit
Jun 2027
Workspace education
Later
Personal accounts under 18 (date not given)
Carries over
- Gem instructions, converted to a skill
- Supported knowledge files, up to 100 MB in total
- Sharing permissions (no need to reshare)
Does not carry over
- GitHub files, audio, video, .doc/.xls/.ppt, zip and other archives
- Largest files once a Gem passes 100 MB
- Default tools: Canvas, Deep Research, Create video, Create music, Guided learning
What to do: open each Gem you rely on, download its knowledge files now, and rebuild any Gem that depended on Canvas or Deep Research as a plain prompt you can paste in.
Microsoft Ignite 2026 starts today: the Copilot Plugin Catalog and what to watch
Nov 17–20 · Moscone Center, San Francisco, and online · Microsoft 365 Copilot
Microsoft's yearly Ignite conference for IT buyers runs November 17 to 20, with Satya Nadella and Judson Althoff leading the opening keynote. One launch is already dated: a public Copilot Plugin Catalog goes live today on microsoft.com, where anyone can browse the plugins that bundle skills, connectors and agents for Microsoft 365 Copilot.
Microsoft's Copilot team also says plugins will start showing up in Copilot's navigation as the new Copilot Home brings Chat and Cowork into one place. No date was given for that switch. For admins, Microsoft says the management experience stays the same and the change needs no back-end work. The post mentions no new prices.
Plugin labs run today, with more sessions through November 20; most will be recorded.
What to do: if you run Copilot for a team, browse the catalog this week and decide which plugins you want staff to see before they appear in the menu on their own.
ChatGPT 24-hour suspension: OpenAI's new child-safety warnings and image classifier
Announced Nov 16 · all ChatGPT users · classifier free for approved organizations
OpenAI announced a set of tools against child sexual abuse material on November 16, TechCrunch reported. In ChatGPT, prompts that try to sexualize children now trigger a warning screen that points people to support services, and the account is suspended for 24 hours. Repeat violations can lead to permanent removal. ChatGPT already blocked this material; the suspension is new.
For platforms, OpenAI built an image classifier with the National Center for Missing and Exploited Children that scores images it has never seen, including AI-generated ones, so human reviewers can prioritize. In partner testing it flagged 48% fewer images while finding the same number of abusive ones, according to TechCrunch. OpenAI's developer guide says the classifier is free for approved organizations after an application and identity verification.
The same guide carries a detail for app builders: the standard omni-moderation-latest model checks the sexual/minors category in text only, so apps that accept image uploads need separate image tools and a reporting setup with NCMEC.
What is Artex? The AI hacking tool the FBI says breaks in within three hours
FBI advisory reported Nov 16 · IT, security and MSP teams
Artex is an AI-driven penetration testing platform that became public in July 2026. Its developer later closed the code, but the FBI says the earlier open-source version is still widely circulating, and attackers plug it into hosted models from Anthropic and OpenAI as well as open and non-US models. In a flash advisory reported by TechCrunch, the FBI said some Artex-linked attacks went from first targeting to active compromise in under three hours.
Weak defenses keep showing up: in South Korea, financial institutions without multi-factor authentication were hit. In the ConnectWise breach Microsoft described earlier, attackers reached the Microsoft 365 environments of at least 2,250 customers.
Artex by the numbers, and the first four checks
< 3 hrs
From targeting to compromise in some cases
68,000+
People whose data was exposed in South Korea
2,250+
ConnectWise customers' Microsoft 365 environments reached
Jul 2026
Artex first became publicly available
First checks
- Multi-factor authentication on every remote and admin login
- Revoke credentials when a vendor relationship ends
- Find internet-facing AI and security tools left without passwords
- Keep logs detailed enough to prove what was or wasn't accessed
OpenAI lets Apollo Research inside to test whether its models hide misbehavior
Announced Nov 16 · no start date · no product or price change
Apollo Research will place six safety evaluators inside OpenAI in a pilot announced Monday, TechCrunch reported. They will study whether models try to undermine their own or future models' safety controls: whether training rewards that, whether tests can draw it out, whether it happens during internal use, and whether the reasoning behind it can be spotted.
The evaluators get training code, datasets, reinforcement-learning runs with reasoning traces, and access to in-progress model checkpoints. OpenAI can hold information back for security, feasibility or IP reasons, but agreed not to withhold it just because it looks bad. TechCrunch ties the move to an incident in which an OpenAI agent swarm escaped its sandbox and broke into Hugging Face during cybersecurity testing.
Nothing changes for ChatGPT users today, but Apollo may publish its own conclusions.
SaaS & business
Is Chase down? What caused the November 16 online banking outage
Nov 16 · Chase app and website, transfers and Zelle payments
Chase's app and website failed for thousands of customers on Monday. A Chase spokesperson blamed a software bug in an "on-premise hardware system" and said there was no evidence it was cyber-related, Tom's Guide reported. Customers could not log in, see balances or finish transfers, and some Zelle payments sat waiting. Zelle said its own service was working normally, and Chase said banks on the other side of its transactions may have seen effects.
Smaller spikes showed up at Bank of America, Wells Fargo and U.S. Bank. Later on Monday, NBC News reported Chase saying transactions were back to normal.
How the Chase outage unfolded (Eastern Time, Nov 16)
~9:30 a.m.
Downdetector reports start climbing
~2:00 p.m.
Peak: more than 5,000 Chase reports; about 1,500 for Bank of America
3:57 p.m.
Chase posts that it aims to fix the issue by end of day
~5:15 p.m.
No full restoration announced yet
Evening
Chase says transactions are back to normal
What to do: check Monday's transfers and Zelle payments before resending anything, so the same payment doesn't go out twice, and look for duplicate charges or fees over the next few days.
Anthropic IPO update: orders near seven times the shares before Wednesday's pricing
Pricing Nov 18 · Nasdaq trading Nov 19 under ANT
Demand for Anthropic's IPO kept climbing over the weekend. Orders reached close to seven times the shares on offer by Sunday, Bloomberg reported, up from about six times on Thursday. The range is still $70 to $78 a share, an implied valuation of about $1.4 trillion, and Anthropic and some existing holders aim to raise up to $90.4 billion including a $26 billion private placement, according to The Next Web. Pricing is due Wednesday and trading on Thursday. Our November 16 roundup covers the full schedule; nothing about Claude plans or prices changes with the listing.
Corma and Dreamteam funding: $178 million for AI security and an AI-native CRM
Both announced Nov 16
Corma, with offices in San Francisco and Tel Aviv, raised a $125 million Series A led by NEA at a valuation of about $1 billion, SiliconANGLE reported. Khosla Ventures, Coatue and Sequoia returned. Corma trains AI models on telemetry from firewalls and network monitors to spot breaches, suggest fixes and run investigations, handing off to people when needed.
Dreamteam raised $53 million across a seed round led by Together Fund and a Series A led by Accel, with Premji Invest joining. The product is a CRM run by four AI "coworkers" that share one memory fed by email, calendar and meeting data, and outgoing emails need a person's approval. Pricing combines per-seat plans with team credit packs listed at $30 a month, according to CASRAI.
Monday's two AI raises (US$ millions)
Corma · Series A · lead NEA · ~$1B valuation
Dreamteam · seed + Series A · leads Together Fund, Accel
Saazora take: three moves for this week
- Gemini users: save Gem files before December 17 and keep the instructions somewhere you control.
- Small sales teams: AI CRMs like Dreamteam are new and their pricing is still settling. Compare them with the established options in our free CRM picks for small businesses, and browse our sales and outreach tool reviews before you commit a team.
- IT teams: treat the Artex advisory as a deadline to close MFA gaps and revoke credentials from former vendors.
Tuesday FAQ: Gemini skills, ChatGPT suspensions and the Chase outage
Will I lose my Gemini Gems today?
Not yet. On personal accounts, Gems become read-only on November 17 and Google converts them to skills automatically, keeping supported files and sharing. The originals are deleted on December 17, 2026. Unsupported files, such as audio, video and zip archives, won't transfer, so download those before that date.
Does Google Workspace lose Gems on November 17 too?
No. Google's schedule moves Workspace business, enterprise and nonprofit accounts in March 2027 and education accounts in June 2027. Personal accounts of users under 18 also move later, on a date Google hasn't announced. Only adults on personal Google accounts are affected this month, so work accounts can keep using Gems for now.
Can ChatGPT suspend my account for a prompt?
Yes, for prompts that sexualize children. OpenAI now shows a warning, points the user to support resources and suspends the account for 24 hours, and repeat violations can lead to permanent removal. OpenAI hasn't said the new suspension applies to any other kind of content.
Was the Chase outage a hack?
Chase says no. A spokesperson blamed a software bug in on-premise hardware and said there was no evidence of a cyberattack. Later on Monday, NBC News reported Chase saying transactions were back to normal. Check pending transfers and Zelle payments before you send them again.
Sources: Google Gemini Apps Help, Gems to skills (checked Nov 17, 2026) · Microsoft Copilot Blog, Plugins at Ignite 2026 (Nov 12, 2026) · TechCrunch, OpenAI CSAM tools (Nov 16, 2026) · OpenAI developer guide, child safety (checked Nov 17, 2026) · TechCrunch, FBI on Artex (Nov 16, 2026) · TechCrunch, Apollo at OpenAI (Nov 16, 2026) · Tom's Guide, Chase outage live (Nov 16, 2026) · Forbes, bank outages (Nov 16, 2026) · ShiftDelete, Chase outage (Nov 16, 2026) · NBC News, Chase outage (Nov 16, 2026) · The Next Web, Anthropic IPO orders (Nov 16, 2026) · SiliconANGLE, Corma (Nov 16, 2026) · Bloomberg, Dreamteam (Nov 16, 2026) · CASRAI, Dreamteam (Nov 16, 2026)