KDDI Breach Hits 12 Million Email Accounts

Japan's second-largest mobile carrier, KDDI, has confirmed a breach affecting customers across five internet service providers, according to Bleeping Computer.

Attackers got into a shared email platform on May 16 by exploiting a zero-day vulnerability (a security flaw unknown to the software vendor at the time of attack) in third-party software. The intrusion went undetected for a month. KDDI discovered it on June 17.

The numbers: 12.2 million email addresses and 7.6 million passwords were exposed across current and former customers of STNet, JCOM, Chubu Telecommunications, NIFTY Corporation, and BIGLOBE. Some passwords were stored in hashed or encrypted form, which makes them harder to misuse. KDDI has not confirmed how many were stored in plaintext.

Attackers with your email address and password can try that same combination across dozens of other services. Think of it like someone copying a key to your front door and then testing it on every flat in your building.

KDDI has since deployed EDR (Endpoint Detection and Response) software to catch future intrusions and has mandated forced password resets through affected ISPs.

What to do: If you use any of these ISPs — or if you reuse passwords anywhere — change those passwords now. Use a password manager to generate unique ones for each account. Enable two-factor authentication on your email wherever possible.

Researchers at the AI Now Institute have published a proof-of-concept attack, dubbed "Friendly Fire," showing that AI coding agents built to catch malicious code can be manipulated into running it instead, according to The Hacker News.

The attack targets Claude Code (Anthropic's autonomous code-execution tool, used via command line) and Codex (OpenAI's equivalent) when either is operating in auto-approval mode. In that mode, the agent approves its own commands without asking the user first.

The method is straightforward: an attacker drops a hidden malicious binary into an open-source library and adds a line to the README suggesting users run a script called security.sh before submitting code. The agent reads the instruction, decides it looks routine, and executes it. The attacker's payload runs on the host machine with no warning.

Both agents, when asked directly whether the library contained hidden instructions, said no.

What to do: Disable auto-approval mode in Claude Code and Codex. Never run these agents unsupervised against code repositories you do not fully control.

What UniFi Connect is: UniFi Connect is a Ubiquiti platform for managing displays, intercom systems, and building access hardware — popular in small offices, schools, and home setups.

What it is: This flaw allows anyone on the same network to inject and execute arbitrary commands directly on the host device, with no authentication required. It is an improper access control vulnerability, meaning the software fails to verify whether a user has permission before carrying out sensitive operations.

Who's at risk: Anyone running UniFi Connect Application version 3.4.16 or earlier on a network where other users or devices are present. Home lab users, small businesses, and schools using Ubiquiti gear are all in scope.

CVSS: 10.0 (Critical) — Patch today. This is a perfect score. Stop reading, go patch.

Root cause: The application does not enforce access controls before processing certain commands. A network-adjacent attacker can send a crafted request that the device treats as a trusted instruction, then execute commands at the system level. There is no authentication barrier in the way.

Attack vector: An attacker on the local network sends a specially crafted request to the UniFi Connect service. The device processes it without checking who sent it, executes the embedded command, and hands the attacker control of the host. No credentials needed.

Recommended actions:

1. Update UniFi Connect Application to version 3.4.20 or later immediately.
2. Restrict network access to UniFi management interfaces to trusted devices only.
3. Review other UniFi products in your environment — Ubiquiti patched six additional critical flaws across UniFi Talk, Access, Protect, and OS in the same release cycle.

The Friendly Fire research above makes one defensive tool directly relevant today: AWS has released the Claude Apps Gateway, described by Help Net Security. It is a self-hosted control plane for organisations using Claude Code (Anthropic's autonomous coding tool) in a team setting.

Instead of each developer holding their own long-lived credentials, the gateway issues short-lived tokens that expire within an hour. It centralises policy controls and enforces tool permissions across your whole team. Organisations managing multiple developers can use it to limit which tools agents can access and revoke access instantly by removing a developer from your identity provider.

This directly limits the blast radius if an agent is manipulated into executing malicious code. Fewer permissions mean less damage. Find it via Help Net Security.

Industry body CREST launched its AI Charter on July 9, with 73 founding signatories committing to nine principles for responsible AI use in cybersecurity, according to Infosecurity Magazine. The principles cover transparency with clients, human oversight of AI decisions, and strict data handling rules. For everyday users, this means security firms that signed up must tell you when AI is involved in their services and how it affects your data. CREST reports that 69% of cybersecurity providers already use AI daily — this charter is the industry trying to get ahead of regulation before regulation gets ahead of them.