PHP2Shell confirmed in attack; Japan's GMO suffers breach

PHP2Shell has a confirmed attack and the first validated RCE documented

The PHP2Shell flaws now have a confirmed attack. Unit 42, the research team at Palo Alto Networks, says the exploitation happens on the internet; a researcher also published the first validated complete intrusion, with the analysis performed by an AI agent.

PHP runs a large part of the corporate web. Unit 42 sees about 22.6 million exposed instances, and servers without the patched versions are a direct target. The company also removed the link between a European government case and this flaw.

Takeaway: Patching base software demands speed: millions of servers stay exposed in the first days.

Sources: JPCERT/CC, @DailyDarkWeb, @0x4D31, @zerodayalpha, The Hacker News

PHP publishes mitigation for new critical remote code execution flaw

A new critical PHP flaw received mitigation guidance. Named CVE-2026-115560, it allows remote code execution. The PHP project says the exploitation conditions are limited and that most deployments are probably not affected.

Industrial vendors that embed PHP, such as Siemens, Schneider Electric and ABB, confirmed affected products. Researchers note that more potent exploits tend to appear over time, and extra days to patch are rare.

Sources: @ImposeCost, @hakluke

FortiSIEM: Canada confirms attacks and Fortinet denies known exploitation

The government of Canada reports ongoing attacks against FortiSIEM, a security monitoring tool from Fortinet. The flaw allows intruders to run commands without a password. Fortinet, in turn, still lists the flaw as having no known exploitation.

Horizon3 published a working exploit on October 20 and there is a public proof of concept. A SIEM holds credentials for much of the network, and such a compromised device greatly expands the reach of an attack.

Sources: @DarkWebInformer, @IntCyberDigest

Malicious sites drop AGENTS.md in the Downloads folder to trick AI agents

Research documents a real attack against coding AI agents. Malicious sites write an AGENTS.md file, a kind of instruction that AI agents read, into the user's Downloads folder. If the agent runs in that folder, it can read the file and obey the attacker.

The guidance is to check the Downloads folder for unknown AGENTS.md files before running agents. Developers who run agents in download folders follow third-party instructions without knowing it inside the corporate machine.

Takeaway: AI agents follow instructions wherever they find them; limit their working directory.

Sources: @WeizmanGal

Brotli fix had to be redone; WAF gains emergency rules

Microsoft redid the emergency .NET fix for the Brotli flaw, a component that decompresses data on the web, because two October 13 fixes were incomplete. Malicious compressed data can corrupt the memory of applications that decompress it.

Crimeflare released firewall rules to detect attempts, active in logging mode by default. The flaw, called HTTP2Shell in some reports, reaches Node.js.NET, Python and proxies that handle compressed data.

Takeaway: An emergency patch applied without retesting can leave the flaw alive, as this case showed.

Sources: Microsoft, The Hacker News, @IntCyberDigest, @BleepinComputer, @DarkWebInformer

Tailscale: 11 flaws remain unfixed, but a patch is on the way

The 11 flaws disclosed in Tailscale remain unfixed, and the company says a fix is on the way. Researcher Erica Windisch published the report without prior notice to the vendor. The tests covered client version 1.104.2 and headscale 0.29.5.

The most severe scenario requires an already compromised device: from it, the attacker gains full access to other servers via Tailscale SSH. Those who use the network only with their own devices reduce the risk, according to experts' reading.

Takeaway: A network that connects servers must be treated as an intrusion path, with least privilege.

Sources: Tailscale, Cyberdione Labs, @IntCyberDigest, @TheHackersNews

KeePassXC 2.8.0 fixes flaw when importing old files

Password manager KeePassXC released version 2.8.0 with security fixes. The main one closes a memory flaw triggered when importing a malicious old file. The update also fixes browser integration issues and Windows security issues.

The attack requires the user to import an old KDB file prepared by the attacker, which limits the reach. Still, password managers hold credentials for the whole company, and keeping the patched version is a basic rule.

Sources: @DarkWebInformer

Exploit for AMD GPU security processor is published

A researcher published an exploit for the security processor of AMD Navi31 cards, the component that protects the firmware. The technique, called Wake & Break, allows manipulation of firmware components. The post cites no fix or advisory from AMD.

The attack requires access to the machine itself, so the direct risk to corporate servers is limited. The publication shows that hardware layers also become a target of security research.

Sources: @bl4sty

Extortion group lists the Liberal Party on a leak site

Data extortion group WhiteArmor listed the Liberal Party (PL) on its leak site. The listing gives no volume or type of data, the party did not confirm the incident, and the claim has not been verified.

Analysts warn that party systems hold membership lists, donors and campaign communications, data useful for phishing and disinformation in an election year. Documents attributed to the PL require caution until verification.

Sources: @DailyDarkWeb

Breach of GMO currency exchange service generates unauthorized trades

Hackers breached a currency trading service of Japanese group GMO and clients found trades in accounts they never made. According to reports, client credentials were stolen and destructive trades hit more than 70 brokerages. GMO asked the Japanese government for help.

The wave of attacks on Japanese companies led Prime Minister Sanae Takaichi to call an emergency meeting with ministers, according to Bloomberg. The case shows that stolen credentials become direct financial loss for clients.

Takeaway: Platforms that move money need limits, alerts and out-of-band validation for unusual trades.

Sources: @AndrewCurran_, @MarioNawfal