Twitxy Saved

IRONSCALES

Verified account This account is verified. Learn more

@IRONSCALES

IRONSCALES is the leading enterprise cloud email security platform combining AI and human insights protecting more than 18,000 global organizations.
Atlanta, GA ironscales.com Joined July 2014
1.3K Following 3.5K Followers 10K Posts

Latest posts 5

IRONSCALES Verified account @IRONSCALES
93% say their incident response plan covers AI agents. 77% say an agent has already acted on malicious, misleading, or unauthorized content. Only 52% have formal policies, controls, and incident response processes in place.

Routinely validate IR plans and training with up-to-date, realistic simulations that include agent-targeted attacks along with human-targeted ones. Make sure training and awareness programs reflect how work gets done in the agentic age, and stress-test the agents themselves through frequent red-teaming.

We saw the same gap with deepfakes last year (99% confident, 85% hit).

Move 4 of 5. Next up: keep deepfake defenses strong.

Read the report: hubs.la/Q04C4pV-0

#IncidentResponse #RedTeaming #AIAgents #AgenticSecurity
IRONSCALES Verified account @IRONSCALES
A technology manager at a coffee and beverage company got a Russian-language alert saying a new device had gained access to his account and a certificate had been issued. There was no login page.

A bold phone number and a blue button dialed the same number, and that was the only remedy offered. A tel: control cannot be fetched, screenshotted or reputation-scored, so the threat sits in the feeling the message creates. The printed IP address and handset model were text in the lure, not telemetry. SPF, DKIM and DMARC all returned none at the receiving hop, and composite authentication passed on the implicit path for a domain with no DMARC record. An empty-anchor tracker carried the recipient mailbox in its path. Microsoft sent it to Junk. The mailbox was quarantined and an analyst closed it as malicious. We have no call data, so what a caller would hear is unknown.

(Link to full teardown in comments)
0:29
IRONSCALES Verified account @IRONSCALES
A Google Premier Partner Badge invitation reached a digital marketing employee at an IT-services company. It greeted them by full name, said their agency was chosen for strong Google Ads performance and was signed Google Partners Team.

SPF passed, DKIM passed and compauth passed, and Microsoft filed it to Junk rather than blocking it. The Accept Invitation button went through a click tracker to a lookalike that swaps the l in Google for a q, registered nine days before the send. Nothing in the message was urgent. It manufactured pride, not panic. We did not observe the landing page, so we make no claim about what it collects. Two mailboxes got the campaign under two different subjects, and both were quarantined. Authentication narrows the field. The spelling of the destination is the evidence.

(Link to full teardown in comments)
0:29
IRONSCALES Verified account @IRONSCALES
A fake GoTo missed-call fax notice carried a 3,206-byte HTML attachment that does nothing when opened. It shows one white card reading Security check, one tap to proceed, and waits.

The destination sits encoded in a data attribute and is decoded only inside a click handler: reverse the odd segments, undo a reversed base64 alphabet, XOR with an 84-character key, then click a programmatic link. A sandbox that renders the file but never taps sees no URL, and the attachment scanner rated it clean. The envelope was sloppier than the file: a 19:48 PM timestamp, a caller number whose tel link pointed elsewhere, and a second attachment telling a different story. Nobody observed the landing page, so the redirect is the verified artifact. Decode handler logic statically, and treat the attachment and the message as one object.

(Link to full teardown in comments)
0:29