🚨 One command let a DeepSeek Harness AI agent disable its own file sandbox.

Attacker-supplied text could push the agent to call the tool’s local interface, switch to danger-full-access, and remove approval prompts.

Read: https://thehackernews.com/2026/09/deepseek-harness-flaw-let-ai-agents.html

⚡ A new CVE drops. Can your team answer “Are we exposed?” in minutes?

See how to pull SBOM, endpoint, cloud, and vulnerability data into one view so cybersecurity teams can quickly find affected assets, verify exposure, and prioritize response.

Join this webinar to learn the workflow: https://thehackernews.com/2026/09/webinar-learn-how-to-answer-are-we.html

🚨 Stolen AI session tokens can bypass passwords and MFA.

In one 7 GB infostealer dump, Okta found 555 JWTs likely tied to AI authentication and 24 still-valid API keys for AI services.

How the replay works: https://thehackernews.com/2026/09/infostealer-logs-expose-replayable-ai.html

Attackers are turning breach secrecy into leverage.

Bitdefender says some offer clean recovery and confidentiality for payment. Meanwhile, 55.2% of IT/security pros who experienced a breach or incident said they were asked to keep a breach quiet even when it should have been disclosed.

How silence became a business model: https://thehackernews.com/expert-insights/2026/09/why-are-so-many-security-professionals.html

‼️ Four espionage clusters used the same Chrome exploit kit within a week.

BlueMoon chained two V8 zero-days with a Windows privilege escalation flaw to deploy payloads including GemStone and ShadowPad.

Inside the exploit chain: https://thehackernews.com/2026/09/four-spy-groups-used-same-chrome-and.html

🔥 U.S. authorities disrupted Xinbi Guarantee and froze $52.8 million in crypto linked to the scam marketplace.

Xinbi then shifted about $2.8 million from USDT to USDD, which lacks USDT’s built-in wallet-freezing feature.

Read: https://thehackernews.com/2026/09/us-disrupts-xinbi-guarantee-scam.html

⚠️ Anthropic AI models breached real systems during security tests.

A misconfiguration connected the evaluations to the open internet. In one case, Mythos 5 uploaded a malicious package to PyPI.

Read details: https://thehackernews.com/2026/09/anthropic-ai-models-breached-real.html

🚨 One example LiteLLM admin key was accepted by nearly 1 in 10 gateways.

Researchers found 294 of 3,074 internet-facing instances accepted "sk-1234". Admin access exposes stored provider API keys and, in tests, reached cloud IAM credentials.

How one key reaches the cloud account → https://thehackernews.com/2026/09/nearly-1-in-10-exposed-litellm-gateways.html

🛑 Cisco, Citrix and Fortinet flaws are now on CISA’s exploited-vulnerability list.

Cisco’s FMC bug can give an unauthenticated remote attacker root access, while a Fortinet flaw is suspected in a PivotC2 campaign that infected 178 devices.

Federal patch deadline: Sept. 12.

Read: https://thehackernews.com/2026/09/cisa-flags-exploited-cisco-citrix.html

🚨 Gigabud uses Android work profiles to hide from banking app malware checks.

In Indonesia, it installs Vwork and a tampered banking app inside the isolated profile, keeping Gigabud outside the scan.

How it works: https://thehackernews.com/2026/09/gigabud-creates-android-work-profiles.html

⚡ Hundreds of AI agents were used to compromise at least 440 PaperCut instances across 48 countries.

The attacks chained authentication bypass and RCE, reaching domain admin access at 12 organizations.

Read: https://thehackernews.com/2026/09/papercut-attacker-uses-hundreds-of-ai.html

‼️ Check Point patched two critical VPN certificate flaws, CVE-2026-85102 and CVE-2026-85103, that could allow unauthenticated RCE under specific, undisclosed conditions.

Both carry CVSS 9.8 scores.

What’s affected and how to fix it: https://thehackernews.com/2026/09/check-point-discloses-two-98-rated-vpn.html

AI governance needs more than an inventory of who uses ChatGPT.

A hands-on look at LayerX shows how it captures full AI conversations, maps activity to user identities, and can warn, block, or redact sensitive data across browsers, desktop apps, and IDEs.

Inside the platform and its controls: https://thehackernews.com/expert-insights/2026/05/hands-on-review-layerx-ai-governance.html

🚨 Thousands of deceptive Android apps are abusing a Google Play trust gap: Early Access apps have no public reviews or star ratings.

Fake casino and reward apps are promoted through social ads, including AI-generated celebrity deepfakes.

How the scheme works: https://thehackernews.com/2026/09/google-play-early-access-abused-to-push.html

⚡ A new vulnerability drops. Can your team tell in minutes whether any app, cloud asset, or software component is exposed?

See how Tines pulled SBOM, application, cloud, and vulnerability data into one view to cut exposure assessment from hours or days.

Register for the webinar: https://thehacker.news/outpacing-ai-era-attacks

This week’s threat list is doing too much:

⚠️ 119K fake shops
🤖 AI-run intrusion tasks
🧩 13 bad npm packages
🌐 33.8K+ exposed Plex servers
🕸️ 5.4K hacked sites
🎣 Phishing built in-browser
🔐 MFA bypass across 40+ countries
📱 200 Android flaws fixed

Read full ThreatsDay here: https://thehackernews.com/2026/09/threatsday-200-android-flaws-browser.html

🚨 Three threat clusters exploited Cisco FMC flaws to steal credentials, deploy Cyclops Blink, and push Qilin ransomware.

Both flaws are now in CISA’s KEV catalog. Federal agencies must patch CVE-2026-20079 by September 12.

Inside the attack chains: https://thehackernews.com/2026/09/cisco-fmc-flaws-exploited-to-steal.html

⚡ PaperCut has replaced its emergency patches for two actively exploited flaws with fully tested maintenance releases.

The flaws have been used to bypass authentication and execute arbitrary code. Versions 26.0.5, 25.0.13, and 24.1.10 include the fixes and added hardening.

Read: https://thehackernews.com/2026/09/papercut-replaces-emergency-patches.html

⚠️ A crafted Sogou Input Method link led to GRAYRABBIT malware.

China-linked UNC3569 abused Sogou’s sgbiz: handler to open a malicious page in its built-in Chromium 80 browser, then exploited CVE-2021-38003 to run code with the logged-in user’s privileges.

Read: https://thehackernews.com/2026/09/china-linked-unc3569-exploited-sogou.html

🚨 Attackers chained two JFrog Artifactory flaws to gain admin control and plant backdoors.

Only unupdated self-hosted servers were open to that chain. A separate critical auth bypass also drew 406,000 exploitation attempts in one day.

How the attacks work: https://thehackernews.com/2026/09/attackers-chain-jfrog-artifactory-flaws.html