aeroplane turbine assembly

Security Highlight: Sixty Seconds in the E&E Bay

Researchers from UC San Diego and Oberlin College have published Design and Implementation of a Physical Implant Attack on the Boeing 737 at the 35th USENIX Security Symposium. The work demonstrates how a small hardware implant, fitted into an existing maintenance connector, can rewrite messages between the aircraft's Flight Management Computer (FMC) and the pilot's Multi-function Control and Display Unit (MCDU) — potentially altering the autopilot route or forging the indicated takeoff weight.

Empirical avionics security research carried out on genuine hardware is rare, and the findings extend well beyond aviation. They show how a component added for electrical robustness can become an attack primitive, and how maintenance interfaces quietly extend a trust boundary.

How a Maintenance Connector Became an Entry Point

The 737 uses a federated avionics architecture in which individual line replaceable units are interconnected by point-to-point ARINC 429 buses. Most are concentrated in the Electronic and Equipment (E&E) bay, reached through an unlocked external hatch below the nose gear that is within reach from the ground. Inside, the researchers identified a maintenance connector wired into two ARINC 429 buses: one carrying display text from the FMC to the captain's MCDU, and one carrying keypresses back. From wiring diagrams and published maintenance videos, they estimate that opening the hatch, seating an implant and closing up again takes around 60 seconds.

Figure 1: Boeing 737 E&E Bay Access Hatch Location (Source: Chris Brady — The Boeing 737 Technical Site http://www.b737.org.uk/)

Their implant fits almost entirely inside the connector housing beneath its dust cap and draws power from the connector itself. Once installed, it can inject keypresses into the FMC and rewrite the text sent to the display, concealing the resulting changes from the flight crew. The scenarios they considered most significant were an executed route change in flight, an altered zero fuel weight and an altered assumed temperature — each feeding calculations that matter during takeoff or cruise.

Figure 2: Examples of selective text modification (left) and full screen override (right) of the MCDU’s default IDENT page (Source: Crow et al., USENIX Security '26.)

photo of two pane displays with flight information

The Bus Driver Attack

Gaining access to a bus is not the same as controlling it. Because ARINC 429 is a single-master design, an override attack was previously thought to require physically unplugging the legitimate transmitter and substituting a malicious one. The researchers defeat that assumption electrically rather than through any protocol weakness.

Figure 3: 737 FMC/MCDU Test Setup (Source: Crow et al., USENIX Security '26.)

hardware testing setup

ARINC 429 transmitters connect to the bus through 37.5 Ω series resistors, which match the characteristic impedance of the twisted pair and limit current during a fault. An implant connected directly to the same bus without those resistors can drive substantially more current and dictate the voltage the receivers observe; by measuring its own drive current, it can simultaneously infer which bits the legitimate transmitter was attempting to send. The result is a complete attacker-in-the-middle capability with no wire splicing, no unit replacement and no interruption of the bus heartbeats that would otherwise raise a maintenance fault. The override was made possible by a feature that was intended to increase the reliability of a bus.

The researchers disclosed to Boeing in April 2020 and validated their prototype in Boeing's own 737 testbed in December 2023. Boeing's published statement confirms it evaluated the FMC design, systems installation and interfaces, and states that the layers of protection present in the system design and operating environment significantly limit the feasibility and risk of real-world attacks.

Threat Modeling Decides What Gets Found

This research produced a result because it questioned an assumption rather than a control. Physical access to an airframe had long been treated as out of scope, on the grounds that an adversary with unrestricted access could cause harm by simpler means, and that airport access controls make such access impractical. Neither argument is unreasonable. But narrowing the model to a bounded case — brief access, by someone with legitimate proximity — brought a real attack surface into view that a control-by-control review would never have reached.

This is why structured threat and risk assessment is mandated in aerospace and defense cybersecurity frameworks. A TARA identifies assets, the threats against them and the attack paths an adversary might follow, and its value lies in exposing the scope decisions that would otherwise remain implicit. Structured assessment pushes toward the question this paper answers: not whether the declared controls work, but whether the boundary was drawn in the right place.

A threat model reflects the conditions in which it was written, and those conditions move. The assumptions that excluded physical access were formed when implants were larger, costlier and demanded specialist knowledge; this work was largely the output of one graduate student with modest funding. Assessments should be revisited as attacker economics shift, not only when a product changes. Keysight's security evaluation services draw on knowledge gained from critical applications across multiple industry verticals and can support design reviews and risk assessments of this kind.

Trust Boundaries in Legacy Communications

The objective is never perfect security but resilience in the presence of the bugs and vulnerabilities any complex system will contain. That is why identifying trust boundaries and modeling information flow matters more than counting vulnerabilities. The central finding here is that access to a maintenance connector violates the assumed trust in a flight-critical bus, and the route change confirmation substantiates it: because the confirm command and its indicator lamp both travel on buses the implant controls, an interlock intended as an independent safeguard is nothing of the kind.

The most transferable design lesson is that subsequent bus implementations harden this trust boundary through better isolation, both physical and logical, rather than layering a control on afterwards. Galvanic isolation, for example, gives every device on the bus — an implant included — an equal and limited ability to influence the signal, so a rogue transmitter can attenuate a legitimate signal but never cancel or replace it. For practical guidance, CISA's Foundations for OT Cybersecurity: Asset Inventory Guidance for Owners and Operators addresses both the segmentation these boundaries depend on and the insecure legacy protocols that make them necessary.

Testing Legacy and Successor Buses

Assessment begins with visibility: asset inventories and data flow models depend on observing legacy signals at the physical layer and decoding them reliably. Traces captured on the Keysight’s DS1060A pattern-based trigger generator in the DS1050A testbench can be decoded in Infiniium using Military and Aerospace Protocol Decoding, covering ARINC 429, MIL-STD-1553 and SpaceWire. Isolating the risky event then matters more than characterizing the average case — the implant activates on recognizing a specific field label, and the pattern matching capabilities of the DS1060A can learn these sequences and fire a trigger. Detection itself is largely an analog problem, best validated by high-resolution acquisition and repeatable stimulus rather than protocol conformance testing alone.

Successor buses shift the problem rather than removing it. MIL-STD-1553, ARINC 629 in its commercial application, and ARINC 664 Avionics Ethernet have evolved to provide better electrical isolation, so evaluating immunity calls for specialist fault injection equipment such as our DS1142A 30A glitch amplifier and DS1070A glitch pattern generator. Today's state of the art also becomes tomorrow's legacy — ARINC 429 dates from 1977, did not begin to be superseded until 1995 and still remains a workhorse of commercial aviation. Assume the same longevity for equipment being designed now, and continually test as new attacks emerge.

What This Means in Practice

For security architects: Physical access deserves a place in the threat model even where it has traditionally been excluded. Ask which components sit inside your trust boundary electrically rather than logically, and whether any safeguard you rely on shares a channel with the asset it protects. Be prepared to work with legacy interfaces, redrawing trust boundaries using OT security patterns.

For hardware engineers: Impedance matching, fault current limiting and similar protections should be assessed for the asymmetry they hand an attacker with bus access. Where a bus carries safety-critical data, galvanic isolation puts every participant on an equal footing and should be considered early, because coupling decisions are very hard to revisit once a fleet is fielded.

For test and validation engineers: Enumerate and characterize maintenance and debug connectors as part of the assessment rather than treating them as out of scope. Capture and decode the legacy signals first, use pattern-based triggering to isolate the sequences that matter and to fire fault injection at exactly those points, then validate physical-layer detection with acquisition that resolves voltage and current anomalies at bit level.

Want more stories like this? Subscribe to the Keysight Device Security Bulletin for monthly highlights on device security trends and practical insights — brought to you by the Keysight device security team.

Related Posts

limit
3