Six Stories Worth a Second Look
Not every story earns a headline of its own, but several developments from the past week deserve attention. This roundup covers a novel electromagnetic eavesdropping technique, a federal sentencing in a SIM-swap fraud ring, an independent audit of Anthropic's AI-driven vulnerability disclosure program, a phishing filter bypass borrowed from AI prompt-injection research, a $10 million U.S. bounty on an Iranian cyber commander, and new reporting tying a Chinese hacking group to the People's Liberation Army.
InjectEave Attack
Researchers publishing ahead of USENIX Security '26 demonstrated a new class of electromagnetic side-channel attack called InjectEave. Unlike earlier techniques that passively listen for stray EM emissions — historically limited to eavesdropping distances under 1.5 meters — InjectEave actively injects a tuned radio-frequency carrier signal into a target device. That injected signal interacts with nonlinear hardware components, such as audio amplifiers, analog-to-digital converters, and switching MOSFETs, which unintentionally "mix" a secret analog signal (like speech) onto the carrier and re-radiate it as measurable EM leakage.
The result is a dramatic range increase: the team demonstrated through-wall and up to 30-meter audio eavesdropping across 11 commercial off-the-shelf devices, including wired headphones from Sony, Dell, and Apple; wireless headphones from UGreen, Philips, and HP; a Flyingvoice VoIP landline phone; and smart fans and lamps. Beyond audio recovery, the researchers outlined an "eavesdrop-synthesize-inject" chain against a VoIP phone — capturing call audio, generating a voice-cloned response, and injecting synthesized speech back into the device's audio path. Because the attack exploits analog hardware nonlinearity rather than a software flaw, the researchers note that conventional digital-side-channel defenses like masking and clock randomization don't apply, making mitigation primarily a hardware design challenge.
SIM Swapper Sentenced
Kenneth Carter, 44, a former AT&T retail store employee in Portland, Oregon, was sentenced to 16 months in federal prison for using his store access to help a criminal ring perform SIM swaps against customers, enabling account takeovers at victims' banks. Carter pleaded guilty in March to one count of conspiracy to commit wire fraud and bank fraud.
According to court documents, the scheme ran from May 2018 to November 2019 and involved at least three other co-conspirators, including a "hacker" who identified victims with online bank accounts, gathered their personal data, and passed it to Carter to reassign the victims' phone numbers. The scheme caused an intended loss of roughly $594,000 across at least three victims — one lost nearly $100,000 — with other transfers stopped by bank fraud-prevention controls before funds could be wired overseas. Carter reportedly received between $1,000 and $2,000 per fraudulent swap. U.S. District Judge Stanley Blumenfeld Jr. also ordered him to pay $99,528 in restitution.
Glasswing Findings Review
Threat-intelligence firm VulnCheck published an independent audit of Anthropic's Project Glasswing, the disclosure ledger tied to Claude's Mythos vulnerability-hunting effort, which launched in April 2026. Of 26,153 total findings logged, VulnCheck found only 202 (0.8%) had actually been fixed after nearly five months, while 245 (0.9%) were withdrawn — meaning more findings were pulled back than confirmed remediated. Roughly 2,096 (8%) had been reported to a maintainer but not yet confirmed fixed, and only about 9.8% of all findings had reached a maintainer's hands at all.
VulnCheck also flagged a severity-rating gap: Claude rated 91.5% of ledger findings with available severity as high or critical, while maintainers who reviewed the same findings rated only about 51–61% that severely. Anthropic has acknowledged that human triage, not model output, is the rate-limiting step in the program, but the numbers underscore that an AI system capable of generating tens of thousands of findings can still be bottlenecked — and potentially misleading on severity — once the findings reach the humans who have to fix them.
Invisible Unicode Phishing Filter Bypass
Microsoft disclosed a large-scale phishing campaign that repurposed ASCII smuggling — a technique first documented in AI prompt-injection research — to evade traditional email security filters rather than to manipulate an AI model. The technique abuses invisible Unicode "tag" characters (in the U+E0000–U+E007F range) that render as nothing to a human reader but still exist in the underlying text.
Instead of hiding instructions for an AI assistant, the attackers inserted invisible tag characters inside financial lure words — splitting "funding" into fragments like "fun[invisible]ding" — to defeat literal keyword-matching filters. Microsoft first detected the signature in early February 2026, when flagged volume jumped from about 21,000 messages to over 1.3 million in a single day; the campaign ran in a high-volume phase for roughly three months, peaking at up to 2.37 million messages per day, before dropping sharply after May 15. A cluster of 148 finance-themed sender domains accounted for about 96% of flagged traffic. Despite the technique's sophistication, Microsoft says Defender for Office 365 still caught over 99% of the messages using other layered signals — sender reputation, IP, and domain checks — rather than relying on the Unicode signature alone.
$10 Million Bounty on Iranian Cyber Official
The U.S. State Department's Rewards for Justice program announced a reward of up to $10 million for information leading to the identification or location of Amir Yaryab, who U.S. officials say leads the Cyber Operations Command within Iran's Islamic Revolutionary Guard Corps Cyber-Electronic Command (IRGC-CEC). Yaryab reportedly holds the rank of brigadier general and previously served as Iran's military attaché to Austria and Slovakia before taking command of Tehran's cyber operations.
U.S. officials allege Yaryab directs multiple IRGC-CEC-affiliated hacking groups — including CyberAv3ngers, Dadeh Afzar Arman, and Mehrsam Andisheh Saz Nik — that have targeted critical infrastructure across defense, energy, financial services, telecommunications, shipping, and travel sectors in the United States, Europe, and the Middle East. CyberAv3ngers previously drew attention for targeting internet-exposed Unitronics programmable logic controllers at U.S. water utilities using default or weak credentials. The reward falls under the legal framework covering foreign-government-directed cyberattacks on U.S. critical infrastructure in violation of the Computer Fraud and Abuse Act.
Chinese Hacking Group QTFY Military Ties
Research firm Natto Thoughts published follow-up analysis expanding on last month's joint FBI/NSA/Cyber National Mission Force advisory about QTFY, a China-linked hacking group attributed to Nanjing Xinjiuwei Network Technology Co. (XJW). The August 26 advisory said QTFY has operated since at least 2018, developing malicious tooling, trading exploits within freelance hacking networks, and targeting U.S. government and critical-infrastructure networks including NASA, the Federal Reserve, the Department of Energy, and the U.S. Senate — with the FBI estimating more than 300 American organizations were affected before a court-authorized operation seized domains underpinning QTFY's QScan and QTRouter platforms.
Natto Thoughts' new reporting traces further connections from XJW to companies called ELEX and Nanjing Lexbell Information Technology, describing ELEX's historical client list as including China's Ministry of State Security, the Ministry of Public Security, and PLA-affiliated entities. Lexbell is described as maintaining military-focused products, PLA-linked leadership, and contracts with the National University of Defense Technology. The DOJ affidavit underlying the original advisory also noted that QTFY members include former PLA personnel who leveraged those relationships to win contracts and subcontracts supporting offensive cyber operations — reinforcing the picture of a private "enabling company" operating in service of Chinese state and military cyber goals.
Key Takeaways
- InjectEave shows electromagnetic side-channel attacks no longer require passive proximity — active RF injection can turn everyday headphones and smart-home devices into eavesdropping targets from up to 30 meters away, and the fix has to happen in hardware, not software.
- A former AT&T employee's 16-month sentence is a reminder that SIM-swap fraud rings still routinely rely on a paid insider inside a carrier's retail network, not just external hacking skill.
- Independent scrutiny of Project Glasswing suggests that generating tens of thousands of AI-found vulnerabilities is not the same as getting them fixed — human maintainer triage remains the bottleneck, and severity ratings from the model may run hotter than what maintainers confirm.
- ASCII smuggling, first identified as an AI prompt-injection risk, has already crossed over into mainstream phishing evasion — a sign that AI-security research findings can be repurposed by attackers targeting traditional email defenses.
- The $10 million bounty on Amir Yaryab signals continued U.S. pressure on IRGC-linked cyber units tied to attacks on Western critical infrastructure, including the CyberAv3ngers persona behind the 2023 Unitronics PLC attacks.
- New reporting on QTFY's corporate lineage reinforces a recurring pattern in Chinese state-linked cyber operations: nominally private "enabling companies" with documented contracts and personnel ties to the MSS, Ministry of Public Security, and PLA.
Sources
- In Other News: InjectEave Attack, SIM Swapper Sentenced, Glasswing Findings Review (SecurityWeek)
- New InjectEave Attack Allows Hackers to Recover Audio Playing on Headphones from 30 Meters (Cybersecurity News)
- Oregon Man Sentenced to 16 Months in Federal Prison for Abusing His Role at Mobile Phone Store (U.S. DOJ)
- The Anthropic Glasswing Receipts Are Starting to Trickle In (VulnCheck)
- ASCII smuggling crosses over from AI prompt injection to phishing evasion (Microsoft Security Blog)
- US offers $10 million for info on Iranian allegedly behind cyberattacks on critical infrastructure (The Record)
- Widening the Circle of Chinese Hacker Group QTFY (Natto Thoughts)
- FBI Disrupts China-Linked QTFY Infrastructure Used to Steal Data From U.S. Organizations (The Hacker News)