FraudMeNot Daily · Issue #10 · Nov 6, 2026
AI faked CEOs to steal $129 million
Plus an Alpine SSH login bypass, 4 robocall numbers to block and a Claude browser-extension risk.
Good morning. Federal prosecutors in Manhattan say three people used AI to sound and sometimes look like real CEOs and big-firm lawyers, then steered employees into wiring more than $129 million. We also have an emergency OpenSSH fix for Alpine Linux, four robocall numbers to block, a critical Apache OFBiz update, and a browser-extension trick that can push an AI desktop agent to run code.
In today's issue: AI CEO impersonation · New FraudMeNot findings · Numbers to block · 3 action steps · Quick hits · Tip of the day
Top story
AI faked CEOs and lawyers for $129M wires
What happened: On Nov 5, 2026, federal prosecutors in Manhattan unsealed charges against Jacob Soall, Sharon Hazan and Guy Sheetrit. According to ABC News reporting on the Justice Department announcement, the three allegedly used AI between July 2024 and February 2026 to impersonate the voices—and sometimes the images—of corporate executives and of attorneys at major law firms. SecurityWeek, citing the Associated Press and the indictment, says they often started with accounting staff, claimed a confidential financial deal, then put a fake "lawyer" on the call who directed a large wire. Prosecutors say the scheme cost companies more than $129 million. Victims named in the reporting include hospital systems in Missouri and California, a Rhode Island bank, a U.K. fintech firm, a Massachusetts mutual fund company and a Florida pet food company. The three were arrested in Spain and the United Kingdom and await extradition. The charges are allegations; no one has been convicted.
Why it matters: A phone or video call no longer proves who is on the other end. An urgent "keep this quiet" request from someone who sounds like the boss is exactly the pressure that makes people skip the usual checks, and the fake lawyer on the same call makes the wire feel approved.
What to do: Our advice: treat any sudden request for a large wire, especially one wrapped in secrecy, as something that needs a second channel. Reach the supposed executive through a number already in your company directory, not through contact details the caller offers. If a "lawyer" joins mid-call, verify that person through the law firm's main switchboard or your own outside counsel before moving money. Report suspected fraud at ReportFraud.ftc.gov and the FBI's IC3.
ABC News → · SecurityWeek → · DOJ fraud news → · Report to the FTC →
FraudMeNot findings
From the FraudMeNot lookup pages
Phone number: (866) 771-4819 is the most-reported number in FraudMeNot's latest phone data that we haven't featured yet: 162 unverified FTC Do Not Call complaints from Oct 1 to Oct 27, 2026, 90% flagged as recorded robocalls. The top reported topic is debt relief (credit cards, mortgage, student loans). Pre-block it now: save it in your contacts as "SCAM – do not answer" and block it, or use your carrier's or phone's call-blocking feature. If it gets through, don't answer and never call it back · See report
Exploited flaw: Apache OFBiz versions before 24.09.08 have a critical code-injection flaw, CVE-2026-83542. FraudMeNot's report, based on the NVD record published Nov 5, 2026, lists a 9.8 score out of 10: a remote attacker with no account can run code on the server. Apache recommends upgrading to 24.09.08. When we checked on Nov 6, CISA had not added this CVE to its Known Exploited Vulnerabilities catalog. If you work in cybersecurity, check whether you run Apache OFBiz and update it. If you're a small business, forward this to whoever handles your IT · See report · Apache announcement · NVD record
AI agent safety: Researchers at Forever Security showed that a browser extension able to run scripts on claude.ai can open Anthropic's Dispatch feature and send prompts that make Claude Desktop install software and run commands on the computer, including approving its own prompts. The Hacker News reported the demo on Nov 6, 2026; the researchers say they have seen no real attacks and that Anthropic is retiring Dispatch for new users. One step (our advice): turn off Dispatch or other remote-control features you don't need, and remove browser extensions that can run scripts on AI websites unless you trust the publisher · The Hacker News
Look it up
More numbers to block today
Report counts are unverified consumer complaints from the FTC Do Not Call data.
(207) 204-5129 Kennebunkport, ME (207) · 159 unverified FTC reports · 85% robocalls · Top topic: government/business/family impersonation
(830) 457-9244 Big Wells, TX (830) · 154 unverified FTC reports · 87% robocalls · Top topic: debt relief
(331) 248-4967 Geneva, IL (331) · 151 unverified FTC reports · 91% robocalls · Top topic: debt relief
Do this today
3 action steps
1. Verify the next urgent wire. If someone who sounds like your boss asks for a quiet wire or video meeting about a confidential deal, hang up and call them back on a number already in your directory. It takes two minutes.
2. Pre-block today's four numbers. Add (866) 771-4819 and the three numbers in the Look it up section to your contacts as "SCAM – do not answer" and block them. It takes two minutes.
3. Run a small business? Ask whoever manages your IT this week: Do we run Apache OFBiz, and is it on 24.09.08? Do any Alpine Linux boxes still allow SSH password logins? Do we install browser extensions that can script AI websites?
Quick hits
- Alpine SSH bug lets long passwords in: OpenSSH released 10.6p2 after a password-authentication bypass that hits some Linux systems built without PAM, including Alpine Linux's openssh-server. Alpine says an attacker who can reach SSH can log into accounts with no password set by sending a very long password, and that upgrading the package does not restart sshd—you must restart it. Temporary fix: set PasswordAuthentication no and restart sshd. (Alpine advisory, OpenSSH 10.6p2, SecurityWeek, The Hacker News)
- 74 state breach filings counted in October: FraudMeNot's October state data-breach report counted 74 filings reported in October 2026 across the California, Washington and Iowa public lists we could read (64, 3 and 7). These are filings, not unique incidents, and a notice is a company's own report, not a finding of fault. Our advice: if you get a breach letter, use contact details from your own statements, not from the message. (Our report)
- Keycloak SAML flaw can take over accounts: Keycloak CVE-2026-117691 is a critical (9.1) SAML HTTP-Redirect binding flaw. FraudMeNot's report, based on Red Hat's record published Nov 5, says an attacker can feed an unsigned SAML message that the server processes, leading to full account takeover of federated users. Fix: upgrade to Keycloak 26.8.2 or later. (Our report)
Privacy & security tip of the day
Call the boss back on a number you already have. Prosecutors say the $129 million AI scheme worked because a voice or video that sounded like the CEO made staff skip the usual checks. Our advice: for any sudden six-figure wire, especially one wrapped in secrecy, hang up and reach that person through a number already in your company directory or on your own contact card. Do the same for any "lawyer" who joins the call—use the firm's main switchboard, not a number from the message. (ABC News, DOJ fraud news)