Skip to content
IIAIT

AI Today · 16 November 2026

Cursor traces its breach to a leaked Fireworks key, and Shopify stores now speak the Personal Agent Protocol

A weekend brief: Cursor explains how attackers got in and what they took, Sierra releases version 1.0 of an open standard for personal AI agents that every Shopify store now supports, Apple opens a US study of how people use AI for health questions, and Google adds a multi-agent research feature to Antigravity.

  1. 01

    Cursor says a key exposed at Fireworks AI let attackers copy agent traces and reach other internal systems

    In an update dated November 13, 2026, Cursor said the October breach started with an access key issued to Fireworks AI, which Fireworks had exposed. The attacker used it to copy a storage bucket holding historical coding-agent traces from users who had not turned on Privacy Mode; Cursor puts this at fewer than 0.05% of users. Some traces contained API credentials, which BleepingComputer reports opened access to further Cursor tools, including archived non-production code, some support emails and credentials customers had shared with support. Cursor says fewer than 0.01% of customers had credentials exposed, that the last attacker activity was blocked on November 1, and that Privacy Mode data, production systems and customer repositories were not affected. This adds the root cause to the malicious Claude Code extension copy reported earlier.

    What it means for you

    If your team uses Cursor, check whether an admin received a list of exposed credentials and rotate those keys, and never paste secrets into support tickets or agent sessions. Where coding agents handle sensitive code, turn on the tool's privacy or zero-retention setting.

  2. 02

    Sierra releases Personal Agent Protocol 1.0, and Shopify switches it on for every merchant

    Sierra published version 1.0 of the Personal Agent Protocol, known as Poppy, on November 13, 2026. The open standard describes how a person's AI agent can find a business, sign in with OAuth and act on the customer's behalf, and it builds largely on MCP and OpenAPI. Sierra says every Shopify store now publishes a Poppy file automatically, Meta's Muse agent checks for it on every site it visits, and more than 100 design partners are involved, including Mercury and Turkish Airlines. Shopify's extension requires the agent to ask the customer before certain actions.

    What it means for you

    If you sell through Shopify, AI agents can now reach your store in a structured way without any work on your side, so check how your products, prices and policies appear to them. Other businesses can review the published specification and implementation kit before deciding whether to adopt it.

  3. 03

    Apple opens a US study of how people use AI to answer health questions, with OpenEvidence

    Apple launched a research app on November 15, 2026 that lets participants ask health questions and receive answers from OpenEvidence grounded in clinical sources such as NEJM, JAMA and Nature, according to TechCrunch. Participants rate the answers and fill in monthly surveys, and can choose to share HealthKit data. The study aims to enroll up to 200,000 people in the US for more than a year; they must be over 21 and use an iPhone 15 or later with Apple Intelligence turned on. Apple says the data is anonymized and will not be used to train AI models.

    What it means for you

    The study should produce large-scale evidence on how reliable AI health answers are in everyday use. Until then, treat AI health answers as a starting point and confirm anything important with a clinician, and check what any health app does with your data before you share it.

  4. 04

    Google adds /teamwork, a multi-agent research feature with independent checks, to Antigravity

    Google Research introduced /teamwork on November 13, 2026, a feature in its Antigravity development tool in which several AI agents work in parallel on a task while separate verifiers, such as proof checkers, test suites and simulators, check their results. Google says scientists and developers can use it with Gemini 4 models now, together with a Deep Review feature for checking documents that includes two free calls a week. It is also offering 10,000 academic researchers six months of free access to its highest usage tier. The performance figures in the announcement are Google's own.

    What it means for you

    The approach of pairing AI agents with automatic checks is worth copying in your own work: ask for results that can be tested, then verify them. If you write technical reports or proposals, the free Deep Review calls are a low-cost way to get a second read, though the feedback still needs your judgement.