Three stories from the past two days. Anthropic lets Claude look into your bank account, OpenAI launches a separate model for security work, and Meta acquires a tool that lets developers run multiple AI agents at the same time.
Claude: Personal Finance in beta, US only
On 5 November, Anthropic announced Personal Finance in Claude. You connect your bank and credit card accounts via Plaid, then ask questions about your balance, transactions and spending. Claude has read-only access: it cannot move, manage or invest money. You log in to your bank yourself, so Claude never sees your password. What Claude does see: the name of the bank, the name, type and last four digits of each account, balances, credit limits and up to two years of transactions. Anthropic does not use conversations in the Personal Finance environment for training, according to the help page. The feature is rolling out to Pro, Max and Ultra, and only in the United States, confirms TechCrunch.
What does this mean for you? In the Netherlands you cannot use it yet. If you work as a financial adviser or at a bank, keep an eye on it: Anthropic says it is exploring how Claude could work with licensed advisers and regulated parties. Note the difference with ordinary chats: on Free, Pro and Max, Anthropic uses those for training by default unless you switch it off. You can read how Claude scores in the GDPR check for Claude in financial advice and the GDPR check for Claude in banking.
GPT-6 Cyber: first for a small group
OpenAI presented GPT-6 Cyber on 5 November, a variant of GPT-6 Astra for authorised security work such as finding vulnerabilities, building exploits and penetration testing. According to CyberScoop, the model found at least 81 new vulnerabilities in Firefox and built exploits for 948 vulnerabilities in the Linux kernel; everything has been reported and patched. As the system card shows, OpenAI classes the model under its Preparedness Framework as "Critical" for cybersecurity, just like GPT-6 Astra. Access initially goes to a limited number of customers in the Daybreak Red programme, with stricter identity checks and extra oversight of risky behaviour.
What does this mean for you? If you work in cybersecurity, you cannot choose the model yourself for now. For the broader picture of AI in your field, there is the guide for cybersecurity, and the GDPR check for ChatGPT in cybersecurity shows what happens to your input.
Meta takes over Conductor
Meta has acquired Conductor, reports TechCrunch. With Conductor, you can set multiple coding agents to work on a project at the same time. Those agents come from other providers, such as Anthropic and OpenAI. The acquisition price has not been disclosed. Conductor's cloud service will shut down on 20 November, after which users' cloud data will be deleted. The other features will keep working until early next year; by then the team wants to come out with a new product.
What does this mean for you? If you use Conductor's cloud service in software development, retrieve what you want to keep before 20 November. If you are looking for another way to work with agents, compare ChatGPT and Claude.
In brief
- Claude reads your bank account in the US via Plaid, read-only and without training on those conversations.
- GPT-6 Cyber finds more vulnerabilities, but for now is only available through Daybreak Red.
- Conductor's cloud service stops on 20 November following Meta's takeover.
- All models and privacy scores are at models and safety.