October 2026 Security Incident Review: PHP, Binance, Cursor, LiteLLM and More

October is Cybersecurity Awareness Month. In October 2026, 15 well-known companies and projects reported a security incident. A list of them went around X, starting with “PHP got hacked” and ending with “happy cybersecurity awareness month”.

It is a good joke, and it holds up better than you would hope. Read each line and you find a different story. Attackers broke into Railway and Binance through their own systems. At Polymarket and Coinbase, they came in through a vendor. On GitHub, they hijacked maintainer accounts. With PHP, attackers exploited the PHP2Shell vulnerabilities.

Hacker News on October 22 summed up the month.

Attackers breached the company directly

Railway. On October 21, Railway found unauthorized access to the internal systems that store customer environment variables. Railway says the attacker may have decrypted some of those variables. Railway paused deployments, which also took down its dashboard, API and CLI for a time. Railway tells all customers to rotate credentials stored before October 21 (disclosure).

OpenCode. On October 17, an attacker got into the hosted OpenCode Console and exposed the AI provider API keys that 1.4% of Console users had linked. The attacker did not reach the CLI or desktop apps. Users woke up to large bills on their OpenAI and Anthropic keys (BleepingComputer).

An OpenCode co-founder said the attacks continued for the next two days.

Binance. On October 29, about $6.4 billion left Binance wallets across 10 blockchains, most of it in about 20 minutes. This is the largest cryptocurrency theft on record. Binance stopped deposits and withdrawals and said it will cover all losses. TRM Labs and Chainalysis attribute the attack to North Korea (TRM Labs).

Cursor marketplace. On October 29, the Cursor extension marketplace served a modified copy of the Claude Code extension as a normal update. Versions 2.1.312 and 2.1.313 were live for about 11 hours. The modified extension added an attacker SSH key and opened a reverse tunnel. Cursor says the update reached at most 5,000 users (Cursor notice).

Anthropic said the attacker did not touch its systems or its official extension channels.

Fireworks AI. On October 6, Fireworks found unauthorized use of internal credentials. A third party read environment variables, including API keys, from evaluation jobs in a small number of customer accounts. Fireworks rotated keys and has found no access to models, datasets or inference traffic (bulletin).

LiteLLM. On October 21, attackers used stolen GitHub tokens from three LiteLLM engineers to run the release workflow. That workflow published LiteLLM 1.107.0 to PyPI with a credential stealer and a backdoor that run each time Python starts. The bad wheels were live for about 1 hour and 40 minutes. We analyzed the payload in Compromised LiteLLM 1.107.0 on PyPI Installs a Credential Stealer and Backdoor.

Capcom. On October 19, Capcom found and blocked unauthorized access to its player account system. The attacker exposed 11.7 million records, almost all of them email addresses and login metadata, and changed the passwords of 196 accounts. Capcom says the attacker did not reach payment card data (Help Net Security).

TiDB Cloud. From September 9 to October 7, an intruder used stolen service account credentials to access PingCAP internal systems. The intruder copied TiDB Cloud account data, source code and a table of encrypted API keys. PingCAP says it has no evidence that the intruder touched customer data stored in TiDB (PingCAP).

SoftBank. On October 7, ransomware encrypted a cloud region of IDC Frontier, a SoftBank subsidiary, and disrupted 495 customer organizations. On October 17, SoftBank blocked an access attempt on its My SoftBank web portal and took the portal offline for a few days (BleepingComputer).

Attackers came in through a vendor

Polymarket. On October 29, a compromised third-party provider served malicious JavaScript to the Polymarket international site. The script drained user wallets. Polymarket paused trading and says it will return $3.9 million to traders (The Block).

Coinbase. On October 28, Amplitude, the Coinbase analytics vendor, exposed contact data for a small group of Coinbase customers. Coinbase says nobody accessed its own systems and customer funds are safe (Decrypt). CZ listed it with three other data leaks from the same day.

Warp. On October 23, Warp found an attacker using a vendor in its infrastructure. The attacker may have seen agent decision logs and telemetry for about 1% of users. Warp has not named the vendor.

Attackers hijacked user accounts

GitHub Actions. On October 8, attackers used hijacked maintainer accounts to add a fake “Security Audit” workflow to 345 repositories. The workflow sent each repository’s Actions secrets to an attacker server (StepSecurity). Nobody reported a breach of GitHub.

Attackers exploited a flaw in the software

PHP. On October 21, PHP released emergency patches for two flaws known as PHP2Shell (CVE-2026-109618 and CVE-2026-109619). The flaws let an attacker run code on a server without logging in. Attackers used them before the patch came out, and CISA added both to its Known Exploited Vulnerabilities catalog (Help Net Security). Attackers went after websites running unpatched PHP. Nobody breached the PHP project.

What these incidents have in common

Six of these incidents put secrets in attacker hands. In Railway, OpenCode, Fireworks AI, LiteLLM, GitHub Actions and TiDB Cloud, attackers took API keys, tokens or environment variables. In four of them (LiteLLM, TiDB Cloud, Fireworks AI and GitHub Actions), the attackers got in with credentials they had stolen earlier.

These steps reduce the damage when a vendor you depend on gets breached:

• Keep secrets short-lived and scope each one to a single job.
• Set spend limits on AI provider keys, so a stolen key cannot run up an unlimited bill.
• Wait a day before you install a new package version. The bad LiteLLM wheels stayed up for under two hours.
• Check what your dependencies run at install time.

Package Manager Guard (pmg) blocks known malicious packages before npm install or pip install runs them, and can hold back versions that are too new to trust.