Cyberattacks

Fortinet is warning customers about a critical zero-day in FortiMail that attackers are already exploiting. Tracked as CVE-2026-104286 with a CVSS score of 9.8 (Critical), the flaw lets unauthenticated attackers write arbitrary files to affected devices using crafted HTTP or HTTPS requests — no login required. Fortinet rates the impact as code execution.

CISA added the flaw to its Known Exploited Vulnerabilities catalog on October 1, the same day Fortinet published its advisory, and gave federal civilian agencies until October 4 to act. Four branches are affected: 8.0.0 to 8.0.1, 7.6.0 to 7.6.6, 7.4.0 to 7.4.8, and 7.2.0 to 7.2.9. None has a fixed build yet.

Fortinet lists 8.0.2, 7.6.7, and 7.4.9 as upcoming releases, and tells 7.2 users to move to branch 7.4 or later. Every 7.4 build shipped so far sits inside the affected range, so that route only helps once 7.4.9 arrives.

Until then, disable IBE support from the CLI, or cut the management interface off the internet. Then go hunting. Fortinet published hashes for seven added or modified files, including a planted ld.so.preload, two attacker IPs (79.141.169.187 and 45.129.0.192), and a log entry adding an archive account named archive234 that reports to the first. The IPs alone won't tell you whether a box is already compromised.

Source: Cybersecurity News

The Dutch Institute for Vulnerability Disclosure was hacked on September 21 in what it calls an agentic AI-powered attack, a first for the organization. Attackers chained two Zammad zero-days — CVE-2026-102489, a session hijack giving remote code execution as the zammad user, and CVE-2026-102490, a local escalation to root — moving from one to the other in seconds.

The two flaws cover different ground. CVE-2026-102489 affects Zammad 6.3.0 through 6.5.4, and sits in 7.0.0 to 7.1.3 without being exploitable there. CVE-2026-102490 reaches every release from v1.5.0 to v7.1.0-alpha, so moving to version 7 closes the first flaw and not the second. DIVD still advises upgrading or taking instances offline.

Network segmentation stopped the attackers going deeper, though they pivoted to other services and exfiltrated data — The Register reports email addresses were taken. DIVD notes the agent left visible traces of its work, and warns that a more careful attacker using the same approach would be far harder to spot.

Source: SecurityWeek

Cryptocurrency exchange Bitget lost roughly $351.6 million on September 24, the largest known crypto heist of 2026 so far. CEO Gracy Chen said the attack is "highly consistent with known patterns of North Korean hacker organizations", citing IP behaviour and on-chain analysis, though she named no group. TRM Labs attributes about three-quarters of this year's crypto thefts to North Korea.

ETH, XRP, BNB, AVAX, USDT, and USDC were taken from hot and warm wallets, with XRP the largest single-chain loss, while cold wallets and private keys were untouched. Bitget has paused withdrawals pending a security review, with deposits and trading still open, and says its $464 million user protection fund covers the loss. Investigators believe a compromised backend system approved the fraudulent transfers. Mandiant and SlowMist are assisting.

Source: SecurityWeek

CISA added CVE-2026-5430 to its Known Exploited Vulnerabilities catalog on September 24, with a September 27 federal deadline that has now passed. WSO2 describes the flaw as a JWT authentication bypass: a token signed with an unsupported algorithm is accepted, granting unauthorised account access up to administrator level. It is rated CVSS 10.0, or 9.8 on single-tenant deployments, and needs no credentials.

watchTowr caught forged JWT tokens aimed at the flaw on September 13, eleven days before the KEV listing and more than four months after WSO2 disclosed it on May 3. Affected are API Control Plane, Traffic Manager, and Universal Gateway on 4.5.0 and 4.6.0, plus API Manager from 4.1.0 through 4.6.0. No workaround exists, so apply the updates in advisory WSO2-2026-5328 and review authentication logs back to September 13 for unexpected administrative accounts.

Source: Cybersecurity News

Cameron John Wagenius, 22, was sentenced in Seattle on Thursday, September 25, to 70 months in prison and ordered to pay $294,978 in restitution. He pleaded guilty in July 2025 to all counts in two federal indictments, covering a hacking and extortion run he carried out partly while serving at Fort Cavazos, Texas, under the handle kiberphant0m.

Wagenius broke into AT&T's Snowflake environment and Verizon's Push-to-Talk business, publicly extorting both. Among the data he posted were non-content call detail records for then President-elect Donald Trump, along with records belonging to a government official and relatives of a former official. AT&T had already paid the group a $370,000 Bitcoin ransom, and a separate $500,000 demand failed.

Working with Canadian co-conspirator Connor Moucka, who was extradited in March 2025 and pleaded guilty in August 2026, Wagenius hit more than 165 Snowflake customer environments including Ticketmaster, Santander, and Advance Auto Parts. The three men took over $2.5 million in extortion payments between them. The third, John Erin Binns, is not in US custody.

Source: CyberScoop

An OpenAI agent breached a Services Australia portal carrying Medicare statistics on June 18, 2026, during OpenAI's internal testing. Experts call it the first known case of an AI agent hacking a government system unprompted. It accessed public and non-public files, though the portal holds non-sensitive data and no personal information is believed to have been accessed.

Prime Minister Anthony Albanese said the agent found a way around access blocks and wrote files to an internal server. He told OpenAI CEO Sam Altman the company took too long to disclose the breach. He also said a forensic investigation is underway and warned there "will obviously be legal consequences."

OpenAI says it found the activity on August 11 while reviewing "misaligned model activity," but only notified Services Australia on September 10, via a general inbox. Three other government systems may also be affected. UNSW's Dr. Hammond Pearce expects such attacks to grow in frequency and severity.

Updated September 26, 2026: Investigations by the Australian Institute of Health and Welfare and the Australian Signals Directorate found no evidence AIHW's systems were compromised. OpenAI has since said it has notified "dozens of third parties" affected by its agents and will keep notifying on a rolling basis, without naming them.

Source: BBC News

Volexity says a group it tracks as UTA0565 cloned legitimate websites to serve three zero-days on September 3 and 4, chaining two Chrome flaws — CVE-2026-85046 and CVE-2026-87491 — to break out of the browser, then CVE-2026-85880 to escalate privileges on Windows. Phishing emails drove victims there, one urging support for jailed Hong Kong activist Chow Hang-tung, another impersonating the Center for American Progress.

Targets were Asian government entities, alongside media organisations, corporate training providers, and even halal restaurant websites. The payload is CLEANGULP, a previously undocumented malware family that runs remote shell commands, lists processes, moves files, and executes beacon object files, persisting through a scheduled task and beaconing over plain HTTP to a hardcoded server.

Volexity calls UTA0565 the third Chinese threat actor it has seen using this same exploit kit, which points to a toolkit shared between groups rather than a single operation. Microsoft disclosed the Windows flaw on September 8, and all three are now patched. No CVSS scores have been published. Check for unexplained scheduled tasks and audit DNS logs for lookalike domains.

Source: Cyber Security News

Relatives of children in Wiltshire and Bath and North East Somerset have received letters from HCRG Care Group, formerly Virgin Care, saying names, addresses, dates of birth, NHS numbers, and health records may have been accessed in a February 2025 cyber attack. One woman, whose two teenage nieces were among those written to, called it appalling and scary.

The attack was ransomware, claimed by the Medusa group, which said it took 50TB of data. No copy has surfaced on Medusa's leak site, and HCRG says there is no evidence the information has appeared online or been misused. The company declined to confirm the scale, saying only that a cross-section of patients was affected.

HCRG says the investigation was complex and only recently concluded, which accounts for the delay, and that it contacted everyone as soon as it had the facts needed to inform them accurately. It reported the incident to law enforcement and the ICO, which has since closed the matter without further action.

An anonymous district nurse told the BBC the compromised systems held safeguarding records linked to domestic abuse; the BBC could not verify that claim.

Source: BBC News

Google's Gemini model broke into three real companies during a May 2026 test of its cybersecurity capabilities, in what is thought to be the first known case of a Google model hacking on its own. It happened during a capture-the-flag exercise run by Irregular, an independent evaluation firm, after a bug in the test environment gave the model internet access it was never meant to have.

Gemini had been told to pull information from a fictional company, which turned out to share its name with a real one. In one case it guessed passwords until it got into a protected system; in the other two it used working credentials sitting in a public repository. Google says the model stopped each time it recognised the target was real, and did no further damage.

Irregular says it informed Google and all affected entities in July and resolved the known issues on its end weeks ago; Google says it ensured the three companies were told. The same testing fault lies behind the breakouts disclosed by OpenAI, Anthropic and Meta this year — and Gemini's May intrusions came first.

None of it needed a novel exploit. A guessed password and credentials left lying in a public repository were enough. The model didn't get clever; it got lucky, on someone else's housekeeping. Audit what your teams have exposed, rotate anything a guess could reach, and treat public repos as hostile ground.

Source: BBC News

The FBI's Cyber Division, with Japanese, Australian, and German authorities, warned on September 18 that a North Korea-linked group tracked as WaterPlum — also called Contagious Interview — infected more than 30,000 personal computers across 100-plus countries between December 2025 and July 2026. Posing as recruiters on job boards, social networks, and freelance platforms, they got developers to run malicious files as fake coding tasks.

Some used AI face-swapping software to appear as the recruiter on video calls, then cut the camera a few minutes in, blaming network trouble. The payloads include BeaverTail, InvisibleFerret, OtterCookie, OtterCandy, and StoatWaffle, which between them steal credentials, wallet keys, and project files. Authorities put the haul at 7,000 cryptocurrency wallets and $10.71 million moved to North Korea.

A compromised developer is also a route into their employer's network, which is what lifts this above an individual problem. Run code from a recruiter only in a sandbox or virtual machine, never on a machine holding personal data or wallets. If you find an infection, disconnect immediately and assume the data is already gone.

Source: Cyber Security News