N022: AI-assisted attack on WorkOS's sales app reached customer Slack excerpts

An AI-assisted attack framework reached WorkOS's internal sales app and its customer conversation excerpts; Langflow's default superuser login gets a 9.8 fix; Wells Fargo will treat AI agents' requests as the customer's own.

Share

N022 · Tuesday 3 November 2026 · 5-min read · Covers 2 Nov 4 AM PT to 3 Nov 4 AM PT, plus dated catch-ups

Jump to: Agent threats & vulnerabilities · Policy, protocols & compliance · Architecture & harnesses · Other big security news · Pro deep dives

Today in 30 seconds

  • WorkOS says an AI-assisted attack framework got into its internal sales app, which held customer contacts and excerpts of customers' Slack, sales-call and support conversations.
  • Langflow, a tool for building AI agents, signed visitors in as superuser by default through 1.12.8; IBM rates that flaw and an unauthenticated code-execution bug 9.8, fixed in 1.13.0.
  • From 24 Nov, Wells Fargo may treat requests sent by a customer's third-party AI tool as if the customer sent them, and may block such tools.

Top story

AI-assisted attack on WorkOS's sales app reached customer Slack excerpts

Found 30 Oct, disclosed 2 Nov 2026 · WorkOS security bulletin · Incident · no CVE · Relevance 9

Attackers used software that automates hacking to break into an internal sales tool at WorkOS, a company that runs sign-in for other software firms, and reach notes on its customers' conversations. WorkOS's bulletin says an AI-assisted offensive framework automated reconnaissance and exploitation against the app, which held CRM contacts, account histories, and excerpts of sales calls, support threads and customer Slack conversations. WorkOS says production systems and application data were not affected; how the attackers got in is not disclosed.

For leaders: Your vendors' sales and support tools hold your team's words; ask whether those apps sit behind SSO with phishing-resistant MFA and are monitored like production.

For builders: Close direct external access to internal CRM and support apps, alert on bulk record reads, and verify out-of-band any request quoting past conversations.

Pro deep dive ↓

Agent threats & vulnerabilities

Langflow through 1.12.8 signs any visitor in as superuser by default; IBM rates it 9.8

IBM bulletin 28 Oct (updated 29 Oct), reported 2 Nov 2026 · IBM Security Bulletin · Vulnerability · CVE-2026-93680, CVE-2026-110186, CVE-2026-110193, CVE-2026-110192 · Relevance 8

Langflow, a popular tool for building AI agents from drag-and-drop blocks, let anyone who could reach a server act as its administrator and run code on it. Server-side automatic login hands every visitor a superuser session (CVE-2026-93680, on by default in 1.12.8) and CVE-2026-110186 allows unauthenticated code execution; a public Metasploit module already lists a server's MCP tools without credentials.

Do this: Upgrade every Langflow instance to 1.13.0 (no workaround exists), turn automatic login off, and keep instances off the internet until then.

Pro deep dive ↓

Also:

Policy, protocols & compliance

Wells Fargo will treat a customer's AI tool's requests as the customer's own from 24 Nov

2 Nov 2026 · Wells Fargo Online Access Agreement update · Final · Effective 24 Nov 2026 · US · Relevance 8

Wells Fargo told customers that if they let an outside AI tool into their accounts, the bank may treat what that tool asks for as their own request. The updated agreement says requests through a third-party AI tool or agent "may be treated as if they came directly from you", makes the customer responsible for the tool's access, and lets the bank block certain AI tools.

Do this: Find agents holding your company's or executives' banking logins and replace shared passwords with the bank's scoped access where offered.

Pro deep dive ↓

Also:

Architecture & harnesses

Claude Code 2.1.314 can stop the agent writing outside the project, and fixes sessions that started without company policy

30-31 Oct 2026 · Claude Code releases · Release · v2.1.314, v2.1.316 · no CVE · Relevance 7 · Disclosure: AgentSec Brief is drafted with Claude

Anthropic's coding agent can now be barred from changing files outside the project folder, and two bugs that silently dropped company rules or the sandbox are fixed. 2.1.314 adds permissions.blockWritesOutsideWorkingDirectories and fixes sessions starting without managed policy when the cached copy expired, and the sandbox skipping worktrees entered with --resume.

Do this: Pin 2.1.316 or later and set both blockWritesOutsideWorkingDirectories and blockReadsOutsideWorkingDirectories in managed settings.

Pro deep dive ↓

Also:

Other big security news

  • Drupal Commerce Core: Unauthenticated access bypass (CVE-2026-118539); store owners should patch per SA-CONTRIB-2026-254 and audit orders (actively exploited).
  • Florida Department of Education: A student-aid site breach may have exposed Social Security numbers of about 4.9 million people (secondary report; FLDOE notice not yet located).
  • Siemens SIMATIC IoT2050: A CVSS 10.0 Node-RED missing-authentication flaw (CVE-2026-58115) is reportedly used to plant backdoors; operators should apply SSA-834709 (actively exploited, per The Hacker News).

One concept

Delegated authority

When software acts for you with your login, the other side cannot tell you apart, so it treats the software's requests as yours. Wells Fargo's new terms make that explicit for AI agents.

Basic mitigation: Give agents their own limited credentials (read-only where possible, spending caps, an expiry date) instead of your password, so you can see, cap and revoke what they do.

Full glossary


AgentSec Brief is published by AM Technology Solutions. Researched and drafted with AI agents, verified against primary sources, and reviewed before publication. Informational only; not legal, compliance or security advice. Verify before acting.

Pro members get the full deep dives with attack diagrams, detection queries and the fix in order, plus the vulnerability tracker, compliance calendar, open-source picks and watch list. Upgrade at https://www.agentsecbrief.com/#/portal/signup