OpenAI paused tool use on its most capable models after an RL agent got past internet restrictions and reached a public chatbot.

The gap was insufficient DNS filtering. Monitoring caught the behavior within 15 minutes, and OpenAI added blocking at two separate layers.

Read: https://thehackernews.com/2026/09/openai-pauses-tool-use-after-agent.html

⚠️ OpenAI shelved GPT-6.1 Astra after safety tests found deceptive behavior and actions taken without permission.

In simulations, GPT-6 Astra created fake identities and delivered malicious payloads to open-source codebases.

What the tests found: https://thehackernews.com/2026/09/openai-shelves-gpt-61-astra-after-tests.html

‼️ Malicious MCP servers can steal OAuth credentials from affected MCP Python SDK clients.

The vulnerability can expose the client secret, auth code, and PKCE verifier, letting an attacker request a valid access token from the real service.

Details → https://thehackernews.com/2026/09/official-mcp-python-sdk-flaw-can-let.html

Source code tops the data sent to external GenAI models in Verizon’s DLP dataset.

Shadow AI is also the third most common non-malicious insider action, up fourfold. Teramind Field CISO Pete Hadjigeorgiou details what an IRM platform should be able to see, prove, and stop.

The 8 checks → https://thehackernews.com/expert-insights/2026/09/what-to-look-for-in-insider-risk.html

Dutch police arrested a 24-year-old in a ShinyHunters investigation.

Police have not publicly named the suspect. Independent reports identify him as Pepijn van der Stap, previously apprehended in 2023 over data thefts and extortions.

Read: https://thehackernews.com/2026/09/dutch-police-arrest-24-year-old.html

🚨 101 malicious npm packages can quietly add WhatsApp accounts to groups without consent.

The PhantomSub cluster abuses Baileys forks and has logged 490,000 downloads, including 116,000 in the last 30 days.

Read: https://thehackernews.com/2026/09/101-malicious-npm-packages-add.html

Six days of hands-on training, then NetWars and SANS@Night talks in the evenings. SANS CDI 2026 in D.C., Dec 14-19.

Details and registration: https://go.sans.org/JoHtqR

Kiteworks fixed a critical flaw discovered during its precautionary shutdown, affecting a capability enabled for under 1% of customers.

The company says there is no evidence of malicious exploitation.

Read: https://thehackernews.com/2026/09/kiteworks-fixes-critical-flaw-found.html

‼️ A new Spectre-v2 attack can leak arbitrary Linux memory by reusing stale CPU branch predictions.

VUSec showed Branch Target Reuse working end to end on modern Intel CPUs, bypassing the Spectre defenses enabled in its test setup.

🔗 Here’s how BTR attack works → https://thehackernews.com/2026/09/new-spectre-v2-btr-attack-leaks-linux.html

⚠️ Russia-linked Star Blizzard has targeted 100+ Ukraine-linked organizations this year, often with fake event invites.

If a target replies, the group may send a password-protected archive containing an LNK disguised as a PDF. Opening it can start a chain that installs the CosmicPulse backdoor.

How it gets in: https://thehackernews.com/2026/09/russias-star-blizzard-targets-100.html

‼️ French tax data theft stayed undetected for seven weeks, until the attacker claimed it online.

Stolen staff passwords opened the door. DGFIP wasn’t monitoring ADER, and one password reset left an active session alive as E-Contact scraping continued for nearly 16 hours.

Read about this: https://thehackernews.com/2026/09/french-tax-data-theft-using-stolen.html

🛑 Citrix NetScaler CVE-2026-88772 can turn 120 crafted DTLS records into root-level shellcode execution.

watchTowr’s new analysis shows how about 174 KB of reassembled data overruns a 35,840-byte scratch buffer, hijacks control flow, and uses mprotect() to bypass NX.

The flaw is already exploited in the wild.

Inside the exploit chain: https://thehackernews.com/2026/09/citrix-netscaler-cve-2026-88772-exploit.html

🚨 A High-severity OpenSSL bug can leak heap memory to a DTLS peer or crash the process, OpenSSL says.

It can happen when the resend timer fires while a larger handshake message is stuck mid-send.

Latest releases fix 13 other flaws: 1 Moderate, 12 Low.

Read: https://thehackernews.com/2026/09/openssl-fixes-high-severity-dtls-flaw.html

⚡ Attackers are exploiting a Citrix NetScaler flaw to gain root access and hide PHP web shells behind .deb and .sig files.

The attacks also deploy SLAPSHOT, a Python tunneler. In at least one case, attackers used it for internal reconnaissance and credential theft.

Inside the attack chain: https://thehackernews.com/2026/09/attackers-exploit-netscaler-flaw-for.html

🚨 CSuite phishing can turn one phish into both Microsoft 365 session theft and remote endpoint access.

ANYRUN traced 351 related sandbox analyses. Some chains deployed ScreenConnect or Action1, and 51% of submissions came from the U.S.

🔗 How the chain works → https://thehackernews.com/2026/09/us-focused-csuite-phishing-steals.html

AI coding agents asked for review screenshots put internal images in public GitHub repos, researchers say.

Researchers counted 13,000+ images at 300+ orgs, including customer billing records. In 93% of cases, they sat in personal accounts.

Read → https://thehackernews.com/2026/09/ai-coding-agents-exposed-13000-internal.html

🚨 ClickFix now accounts for 52% of Push Security detections.

Four in five intercepted ClickFix payloads are reached through search engines, using compromised sites, malvertising, and SEO poisoning to get users to run malicious commands themselves.

🔗 Inside the six browser attack techniques shaping 2026 → https://thehackernews.com/2026/09/know-your-enemy-browser-based-attack.html

🚨 Attackers are abusing ChatGPT Custom GPTs to infect users with RAT malware.

Sponsored Google results lead to attacker-created GPTs, then to Google Sites pages with fake Cloudflare CAPTCHAs. Huntress says at least 40 users were infected.

Learn how this ClickFix chain works: https://thehackernews.com/2026/09/attackers-abuse-chatgpt-custom-gpts-to.html

‼️ WARNING - Cisco says attackers are exploiting a critical SD-WAN Manager authentication bypass.

CVE-2026-76504 lets a remote attacker access the Manager API as admin without credentials. Cisco has fixes, no workaround, and log checks for suspicious activity.

Read: https://thehackernews.com/2026/09/cisco-warns-of-attackers-exploiting.html

⚠️ Attackers are using phishing emails to turn legitimate MSP360 installs into a path for ScreenConnect.


Microsoft says the dual-RMM setup gives them redundant remote access, then supports tool delivery, information collection, and credential-access activity.


Read: https://thehackernews.com/2026/09/attackers-abuse-msp360-to-deploy.html