AI
Japan’s APT threat landscape
Three APT groups are actively targeting Japan. Chinese groups (Panda) exploit zero-days in IT asset management and endpoint management tools widely deployed in Japanese enterprises. Russian groups run living-off-the-land attacks using PowerShell and WMI, then layer in disinformation campaigns through social media. DPRK groups go after financial institutions and cryptocurrency exchanges. The DM Bitcoin breach illustrates the method: a fake recruiter on LinkedIn sent a malicious Python script disguised as a coding test, stole session cookies, tampered with transactions, and walked away with 48.2 billion yen. That money funded the regime.
“the security tools we trust and use can themselves became the entry point of our attacks.” — Yuma Kurogome
Why conventional pentests leave critical gaps
“simulation alone is not enough.” — Yuma Kurogome
Standard pentests run on fixed scenarios: outside-in web exploitation, AD privilege escalation from a compromised endpoint, insider-threat file exfiltration. Each scope is useful. None covers physical intrusion. Real adversaries cross that boundary — a malicious third-party visitor, a fake hire loading malware the moment their workstation arrives. When the attack surface spans physical and cyber domains simultaneously, a test that stays inside the network perimeter misses the entry point entirely.
On-site operations: badge forgery and hardware implant deployment
Before touching a keyboard, the team watches employees at building restaurants to map dress codes, badge designs, and door habits. At smoking areas, where targets linger and outsiders blend in, an operator closes within range and copies the IC card: Flipper Zero RFID cloning for badge bypass (18:43). Post-entry, a Hardware implant kit: ScreenCrab, PacketSquare, KeyCloc, Q70 recorder (19:33) deploys fast. ScreenCrab taps conference room HDMI; PacketSquare intercepts Ethernet at printers; KeyCloc logs wired keystrokes; a Q70 voice recorder hides under desks on magnets, emits no radio signal, and holds 450 hours on a charge. All devices feed a Raspberry Pi that exfiltrates over LTE, traffic indistinguishable from an employee’s smartphone.
Remote operations: malware delivery, C2 evasion, and lateral movement
EDR blocks Python and Node installers by flagging the install event. The fix is portable runtimes: executables that run without an installation step, so no install event fires and no one asks why a back-office machine is running a script engine. Malware written in Python or Node also travels well — change the C2 address and redeploy. For C2, connections to unfamiliar servers trigger alerts; connections to AWS, Azure, Slack, or Discord do not. Kurogome’s team embeds commands and results inside routine outbound traffic to services the target already uses.
Portable Python/Node malware with legitimate-service C2 (28:10)
SSH keys pulled from compromised endpoints then map the org’s GitHub access. From there, lateral movement reaches every repository that key can touch.
GitHub backdoor implantation and supply chain compromise
SSH keys harvested from compromised endpoints give access to internal GitHub repositories containing CI/CD configs and platform documentation. Inserting a backdoor as a clean commit would look suspicious, so the method hides it inside noise: delete many files in one commit, disguised as human error, to flood the diff. Then restore those files with backdoor code embedded. Reviewers see a large, boring recovery commit. To defeat GitHub’s code search, the backdoor uses homograph Unicode characters — visually identical to Latin letters in existing variable names, but returning no results when defenders search for the string.
GitHub diff-flooding backdoor insertion (30:03)
Three principles of advanced scenario design
Kurogome distills 25 days of undetected operation into three principles. First: simulate adversary decision-making, not techniques. Serious attackers avoid known TTPs, so ask what a defender would consider normal, then use exactly that. Second: cross boundaries. Each physical-to-cyber or cyber-to-physical pivot opens new attack surface that single-domain tests never touch. Third: build redundancy. Identify critical points, physical entry, initial C2, and design the chain so one failure does not stop the objective.
“advanced pen testing is not merely a technical exercise.” — Yuma Kurogome
Q&A
Are there locally-based Japanese APT groups doing financially motivated attacks like ransomware, and do you see attacks on industrial infrastructure? No locally-based Japanese APT groups — the ransomware actors targeting Japan are Russian-based cybercriminals, not government-backed; factory infrastructure is out of scope for this engagement but Rich Security has used fuzzing to find factory vulnerabilities in other engagements. ▶ 35:31
From a defender’s perspective, what are one or two points in the engagement where defenders could have detected and stopped the attack? The biggest gap is the space between written security policies and actual enforcement — policies may prohibit unauthorized devices but attackers find and slip through wherever that gap exists. ▶ 37:21
Notable Quotes
the security tools we trust and use can themselves became the entry point of our attacks. Yuma Kurogome · ▶ 6:24
simulation alone is not enough. Yuma Kurogome · ▶ 9:00
advanced pen testing is not merely a technical exercise. Yuma Kurogome · ▶ 34:12
attackers consistently ask what would be allowed that what would be normal uh what the path of le resistance Yuma Kurogome · ▶ 32:07
Key Takeaways
- Deploy Flipper Zero and passive hardware implants before touching a single keyboard — physical access is the stealthiest initial vector.
- Embed C2 channels inside Slack, Discord, or cloud services already trusted by the target to defeat network-layer detection.
- Simulate adversary decision-making, not just known TTPs — ask what a defender would allow, then use exactly that.
About the Speakers
Satoki Tsuji
Cybersecurity Enthusiast, CTF Player and Bug Hunter. Contributed to the organization of SECCON CTF, took the stage at AVTOKYO2020/2023/2024, Security Analyst Summit 2024, Hack Fes. 2024, and competed in the DEF CON CTF Finals. Renowned for uncovering and reporting vulnerabilities in web services and softwares including Google and Firefox.
Yuma Kurogome
Yuma Kurogome is the Chief Technology Officer at Ricerca Security, a Japanese firm with a proven track record in offensive security. He joined Nippon Telegraph and Telephone Corporation (NTT) in 2017, where he worked on R&D in malware analysis and detection. Since 2021, he has been responsible for the overall technical leadership of Ricerca Security, overseeing initiatives such as vulnerability assessments, penetration testing, and red team operations. He also established and continues to operate a dedicated research base focused on fuzzing. He provides security advisory services to both domestic and international clients — including Fortune 500 companies — as well as to defense and government agencies.