Daily Summary for 2026-10-18

Introduction to the show: ident: Security Radio. Generated commentary on the latest security and cryptography papers.

Nadia: It's the eighteenth of October, twenty twenty-six, and this is the day's research.

Elias: 6 new papers came out today.

Nadia: I'm Nadia.

Elias: And I'm Elias, here with Priya.

Nadia: We'll take the day in one pass, then pull out the papers we're staying with.

Elias: Start us off with Forensic Analysis of the HackRF One with PortaPack Mayhem Firmware.

The summary: Nadia: It is the eighteenth of October, twenty twenty six. Today we focus on securing large language model agents because their interaction with external tools and sensitive data presents major security failures.

Elias: Researchers are looking at a dual-boundary architecture described in SecureClaw which separates authorization from runtime execution.

Priya: This prevents unauthorized external actions and plaintext data exposure during operation.

Nadia: This work is important because it addresses both boundaries simultaneously offering a path to zero attack success rate on certain benchmarks while maintaining task utility.

Elias: Furthermore we need to consider how fine-tuning methods can inadvertently weaken safety alignments.

Priya: That is what Few-Shot Truly Benign DPO Attack for Jailbreaking LLMs explores by showing benign preference pairs can suppress refusal behavior across various model sizes.

Nadia: The security of agent skills needs attention.

Elias: Towards Secure Agent Skills maps the attack surface across phases.

Priya: Structural flaws like no data-instruction boundary are severe threats.

Nadia: That requires more than incremental fixes to solve them.

Elias: It's My Agent, Not Me shows how roles influence sensitive information sharing.

Priya: This context informs broader privacy considerations on LLMs.

Nadia: On the Privacy of LLMs studies attack sensitivity to model scale and data repetition.

Elias: SecureClaw highlights unauthorized external actions and plaintext exposure risks.

Priya: Few-Shot Truly Benign DPO Attack shows fine-tuning can weaken safety alignment.

Nadia: Even benign preference pairs can lead to harmful outputs in Direct Preference Optimization.

Elias: That is a significant finding regarding the training process.

Priya: Towards Secure Agent Skills defines a threat model across the Agent Skills lifecycle.

Nadia: It systematically studies the framework's security by defining its taxonomy.

Elias: It's My Agent, Not Me explores how AI delegation affects privacy boundaries.

Priya: This study looks at communication roles influencing information sharing privacy.

Nadia: On the Privacy of LLMs provides a unified threat model for LLM privacy issues.

Elias: It systematically studies attacks like membership inference and backdoor sensitivity.

Lucky paper: 2610.20647: Nadia: Today we are looking at Forensic Analysis of the HackRF One with PortaPack Mayhem Firmware.

Elias: This study establishes, at source-code level, where forensically relevant data reside on the microSD card of a device running Mayhem v2 point 0.one and in what format they are written and how to interpret them.

Priya: The HackRF One is an open-hardware lowcost SDR that records radio signals digitally as in-phase/quadrature IQ samples, and the PortaPack interface board lets it become a stand-alone tool with its own touchscreen, battery, and microSD card.

Nadia: The most important forensic components are the microSD slot, the RTC backup battery, the SPI flash holding the firmware, and that micro-USB port.

Elias: Mayhem uses ChibiOS/RT two point six.eight real-time operating system and is distributed under a GNU General Public License GPL.

Priya: Since version two point zero.zero some applications are kept on the SD card as APPS files, and the device’s own memory has no file system for user data, so all persistent data is on that microSD card.

Nadia: The study provides an artifact catalogue for Mayhem v2 point 0.one where every artifact is linked to the source-code file that writes it.

Elias: Key artifacts include SETTINGS/.ini files which are re-created at every exit from the application, giving information on when an application was opened and closed.

Priya: The CAPTURES/ directory stores radio captures as headerless interleaved int16 I/Q data files where duration is calculated as size divided by four times sample rate.

Nadia: Deleted data can be detected by the FAT32 setting the first byte of a file’s directory entry to 0xE5 and marking its clusters as free.

Elias: Trimmed copies of files are written to a-tmp file when IQ Trim runs, which provides evidence that a capture was trimmed.

Priya: The study details how file dates are formed through the DATE FILEFLAG file when the backup battery that powers the real-time clock RTC is absent, and an inference of the time zone exploits different rules for Windows and Linux write times.

Nadia: They follow a six-stage workflow starting with acquisition and integrity using MD5 and SHA-two hundred fifty-six hashes on a read-only copy.

Elias: File system analysis involves examining FAT32 clusters, including unallocated space where deleted data may be found.

Priya: The timeline is generated from mactime and the file list, interpreting timestamps according to the system that wrote them, taking into account the state of the RTC battery.

Nadia: Deleted IQ data can be recovered contiguously from free cluster ranges between neighboring captures; for example, a final part of a deleted IQ capture was recovered and six captures were attributed to a single rolling-code on-off keying OOK transmitter sharing one thirty-two-bit identifier.

Elias: The study provides a workflow for decoding remote-control signals in the captures by comparison with an independent decoder called rtl four hundred thirty-three.

Priya: For a C16 capture, you read the center frequency fc and sample rate fs, using the envelope, PSD, and spectrogram to determine if a signal is present and how it is encoded.

Nadia: If the envelope shows only two levels on and off and consists of pulses, the signal is classified as OOK.

Elias: The line code is identified by looking at which lengths of pulses and gaps occur next to each other in pulse-width modulation PWM or Manchester coding.

Priya: The results were validated against an independent decoder called rtl four hundred thirty-three showing that the Manchester decoding was consistent with the independent tool on one hundred fifty-four of the one hundred sixty-three frames.

Nadia: The findings lead to recommendations for practitioners, such as removing the SD card before switching the device on and imaging it as a whole disk rather than only a partition.

Elias: It is recommended to keep the hash report of the acquisition tool together with the image as proof that the image has not changed since acquisition.

Priya: The rules are limited to Mayhem v2 point 0.one and a single SD card, with future work suggesting differential analysis with controlled sets of actions on the same device.