More than 86,000 Fortinet firewalls are compromised across 194 countries, hackers just demonstrated 32 zero-day exploits at Pwn2Own Ireland, and a critical Atlassian vulnerability sits unpatched on thousands of servers. All of this happened in the span of a single week.
Security teams are not dealing with three separate stories. They're looking at a week where old attack techniques met new AI-powered speed, and defenders are falling behind on all three fronts.
The FortiBleed Campaign: A Global Firewall Takeover
The FBI and U.S. Secret Service issued a joint advisory Tuesday warning that the monthslong FortiBleed credential-harvesting campaign has escalated. Threat actors are not just stealing credentials from Fortinet FortiGate devices. They're locking legitimate administrators out of their own firewalls by changing passwords and creating new accounts to maintain persistent access.
More than 86,644 devices are confirmed compromised, according to data from SOCRadar. The campaign reached 194 countries and touched organizations in all 16 U.S. government-designated critical infrastructure sectors, including energy, healthcare, food and agriculture, and defense.
The operation relies on a custom Golang-based sniffer tool that intercepts authentication traffic from internet-facing FortiGate devices. Exfiltrated password hashes were fed into a GPU-accelerated cracking cluster, and the resulting credentials were sorted, validated, and prioritized based on the revenue and network structure of the target.
More than half of the compromised devices are in India, the United States, Taiwan, Mexico, and Turkey, according to researchers at CloudSEK. The FBI advisory links the campaign to INC, Lynx, and Payload ransomware operations.
Fortinet said in a June advisory that the hackers are reusing credentials stolen from prior incidents and then employing brute-force techniques against devices without multifactor authentication. The company urged customers to enable phishing-resistant MFA immediately. Too many did not.
Pwn2Own Ireland 2026: 32 Zero-Days in One Day
While FortiBleed was making headlines, the Pwn2Own hacking competition in Dublin wrapped up its first day with 32 unique zero-day exploits demonstrated. Trend Micro's Zero Day Initiative awarded $388,500 to researchers who compromised Samsung Galaxy S26 phones, Oracle's Autonomous AI Database, Home Assistant Green smart home hubs, and a Lexmark printer running Doom.
Day two added more wins. A team from KAIST Hacking Lab remotely breached the Samsung Galaxy S26 with one unique bug and two collisions. Researchers from Doyensec and multiple other teams exploited Home Assistant Green with chains of five or six bugs. The AI Infrastructure category saw successful exploits against Dynamo and Chroma.
The speed is the story. Researchers are not just finding one vulnerability. They're chaining five or six together to reach remote code execution, and AI-assisted tooling is accelerating the process. Google's Threat Intelligence Group reported that vulnerability disclosures have doubled from 5,045 to over 10,000 per month, with AI-assisted approaches finding proportionally more high-risk flaws that lead directly to remote code execution.
The AI Infrastructure category at Pwn2Own drew particular attention. Oracle's Autonomous AI Database was exploited by multiple teams using chains of bugs, including a seven-bug chain ending in a use-after-free and type confusion. Dynamo, an AI serving platform, was also compromised. These results suggest that the AI tools enterprises are rushing to deploy are themselves becoming attack surfaces before security teams have fully vetted them.
The Pwn2Own results confirm what the FortiBleed campaign already demonstrated: defenders are patching slower than attackers are exploiting.
Atlassian CVE-2026-21589: Patch Now, No Exceptions
The Cyber Security Agency of Singapore issued an alert Wednesday covering CVE-2026-21589, a critical path traversal vulnerability in multiple Atlassian Data Center products. The flaw carries a CVSS v4.0 score of 9.3 out of 10.
The vulnerability affects Bitbucket, Confluence, Jira Software, Jira Service Management, Bamboo, Crowd, Crucible, and Fisheye Data Center products. An unauthenticated remote attacker can access specific files within the web application root directory, potentially exposing sensitive configuration files, credentials, and tokens.
Atlassian has released fixes, and no action is required for cloud customers. The problem is the thousands of self-hosted deployments that have not yet applied the update. The CSA alert is explicit: update immediately.
This is not a theoretical risk. Attackers are actively scanning for unpatched instances, and the exposed credentials and tokens would give them a direct path into internal development and project management environments.
AI-Powered Attack Chains: The New Normal
The Aviatrix Threat Research Center published an analysis this month (also covered in our cybersecurity desk) covering a September attack on the Dutch Institute for Vulnerability Disclosure. The attackers used automated AI agents to chain two Zammad vulnerabilities, CVE-2026-102489 and CVE-2026-102490, both rated 9.4, to achieve session hijacking, remote code execution, and root access within seconds.
The Dutch institute, which coordinates vulnerability disclosure for researchers, was itself compromised through the very kind of ticketing system it uses to manage security reports. The attackers established command-and-control channels, performed lateral movement across cloud infrastructure, and exfiltrated volunteer personal information and internal research data.
A separate campaign tied to the Tren de Aragua criminal network pulled off an ATM jackpotting operation that stole $40.73 million from banking infrastructure. Truffle Security found 543,699 valid credentials exposed in GitHub repositories, many despite existing security measures.
These incidents show that AI tools are lowering the barrier to entry for sophisticated attacks. Organizations that have not adopted zero-trust segmentation, egress security controls, and anomaly detection are operating with blind spots.
What This Means for Security Teams
The three threads of this week's news connect in one uncomfortable way: attackers are faster, better armed, and more organized than the defenses they're facing.
FortiBleed has been running for months, yet 86,000 devices are still compromised. Pwn2Own researchers found 32 zero-days in a single day, and more will be exploited in the wild before vendors ship fixes. The Atlassian vulnerability is patchable, but history says many organizations will wait days or weeks, long enough for attackers to find them.
The practical steps are not complicated. Patch the Atlassian products immediately if you run them on-premises. Audit FortiGate devices for unauthorized admin accounts and reset all credentials. Enable phishing-resistant MFA everywhere it is supported. Segment cloud workloads so that a single compromise does not become a lateral movement path.
The tools to do these things exist. The question is whether organizations will deploy them before the next advisory lands, or after the next breach.