The 2026 SANS survey lands on a reassuring finding: a trained, skeptical human is still the best defense against AI-enabled attacks. Sit with the rest of the numbers and a harder pattern shows up. Skepticism runs on time and discomfort, and an AI-shaped attack is built to remove both, on both sides of the firewall.
Two guys phoned a phone shop, said they were support, and walked out with 365,000 customer records – 41,359 with payment data attached. The Garante fined Wind Tre €1,715,600.
Excerpt: For two days in July, any Instagram user could pull another person’s public posts into an AI-generated image by @-mentioning them, with the setting enabled for everyone by default. Meta withdrew the option after backlash. The interface disappeared. The assumption underneath it, that a person’s public content is reputation exposure rather than attack material, did not.
Phishing is the single most common way into a breached company right now, and it has never been cheaper to run against you. So how did breached organizations respond? By cutting their planned security spend. Same cost-cutting instinct on both sides of the table, opposite blast radius, and only one side collects.
60.000 chats. 24 hours. DPO disabled. Italy laughed at a chatbot that recommends Pepsi for organic food. The part nobody is laughing about is what ‘alignment disabled’ means when a small, cheap, on-prem model lands in the wrong hands. A technical breakdown of provenance, scope, and how to smell-test the fuffa.
France built Tchap as the secure alternative to WhatsApp for 825.000 civil servants. Mandatory since September 2025, sovereign infrastructure, end-to-end encryption on private messages. On June 7, ANSSI detected a breach. The entry point was not a zero-day. It was one public servant account, social-engineered into yielding access.
Banks are warning employees about spear phishing again. The checklists are familiar: verify the sender, use a second channel, follow approval processes. Good hygiene. But the warning itself encodes an assumption that is quietly wrong, that the threat is still mostly about email, and mostly about obvious manipulation. It is not.
Man-in-the-middle didn’t die. It changed altitude. EvilGinx2, Tycoon 2FA, Rockstar 2FA are MITM proxies wearing a phishing kit costume, and 80% of what they capture is corporate. Full breakdown, defensive code, and the angle awareness training keeps missing.
DragonForce spent two weeks inside a major US services firm, routing command-and-control traffic through Microsoft Teams relay servers. No alert fired. The monitoring stack saw legitimate Microsoft infrastructure and moved on. This is what happens when attackers stop breaking things and start borrowing them instead.
138 campaigns in one week, and one of them cloned a university login portal with a kit the same operator already ran against other atenei. The attacker industrialized phishing for free. Here’s the angle nobody’s pricing in, plus three defensive tools you can run today.
