Wie man ein KI-Modell stiehlt, ohne tatsächlich etwas zu hacken

Teilen:

Artificial intelligence models can be surprisingly stealable—provided you somehow manage to sniff out the model’s electromagnetic signature. While repeatedly emphasizing they do not, in fact, want to help people attack neural networks, researchers at North Carolina State University described such a technique in a new paper. All they needed was an electromagnetic probe, several pre-trained, open-source AI models, and a Google Edge Tensor Processing Unit (TPU). Their method entails analyzing electromagnetic radiations while a TPU chip is actively running.

“It’s quite expensive to build and train a neural network,” said study lead author and NC State Ph.D. student Ashley Kurian in a call with Gizmodo. “It’s an intellectual property that a company owns, and it takes a significant amount of time and computing resources. For example, ChatGPT—it’s made of billions of parameters, which is kind of the secret. When someone steals it, ChatGPT is theirs. You know, they don’t have to pay for it, and they could also sell it.”

Theft is already a high-profile concern in the AI world. Yet, usually it’s the other way around, as AI developers train their models on copyrighted works without permission from their human creators. This overwhelming pattern is sparking lawsuits and even tools to help artists fight back by “poisoning” art generators.

“The electromagnetic data from the sensor essentially gives us a ‘signature’ of the AI processing behavior,” explained Kurian in a statement, calling it “the easy part.”  But in order to decipher the model’s hyperparameters—its architecture and defining details—they had to compare the electromagnetic field data to data captured while other AI models ran on the same kind of chip.

In doing so, they “were able to determine the architecture and specific characteristics—known as layer details—we would need to make a copy of the AI model,” explained Kurian, who added that they could do so with “99.91% accuracy.” To pull this off, the researchers had physical access to the chip both for probing and running other models. They also worked directly with Google to help the company determine the extent to which its chips were attackable.

Kurian speculated that capturing models running on smartphones, for example, would also be possible — but their super-compact design would inherently make it trickier to monitor the electromagnetic signals.

“Side channel attacks on edge devices are nothing new,” Mehmet Sencan, a security researcher at AI standards nonprofit Atlas Computing, told Gizmodo. But this particular technique “of extracting entire model architecture hyperparameters is significant.” Because AI hardware “performs inference in plaintext,” Sencan explained, “anyone deploying their models on edge or in any server that is not physically secured would have to assume their architectures can be extracted through extensive probing.”

Quelle

Kommentar verfassen

Deine E-Mail-Adresse wird nicht veröffentlicht. Erforderliche Felder sind mit * markiert

lade-bild
London, GB
11:04 am, Juli 11, 2025
Wetter-Symbol 28°C
L: 27° | H: 30°
wenige Wolken
Luftfeuchtigkeit: 45 %
Druck: 1021 mb
Wind: 2 mph SSE
Windböe: 6 mph
UV-Index: 0
Niederschlag: 0 mm
Wolken: 13%
Regen Chance: 0%
Sichtbarkeit: 10 km
Sonnenaufgang: 4:56 am
Sonnenuntergang: 9:15 pm
TäglichStündlich
Tägliche VorhersageStündliche Vorhersage
Today 10:00 pm
Wetter-Symbol
27° | 30°°C 0 mm 0% 8 mph 47 % 1021 mb 0 mm/h
Tomorrow 10:00 pm
Wetter-Symbol
18° | 30°°C 0 mm 0% 9 mph 65 % 1018 mb 0 mm/h
So. Juli 13 10:00 pm
Wetter-Symbol
17° | 27°°C 0 mm 0% 7 mph 73 % 1014 mb 0 mm/h
Mo. Juli 14 10:00 pm
Wetter-Symbol
20° | 29°°C 0 mm 0% 14 mph 71 % 1017 mb 0 mm/h
Di. Juli 15 10:00 pm
Wetter-Symbol
15° | 27°°C 0 mm 0% 13 mph 71 % 1021 mb 0 mm/h
Today 1:00 pm
Wetter-Symbol
29° | 29°°C 0 mm 0% 3 mph 42 % 1021 mb 0 mm/h
Today 4:00 pm
Wetter-Symbol
30° | 31°°C 0 mm 0% 5 mph 31 % 1019 mb 0 mm/h
Today 7:00 pm
Wetter-Symbol
28° | 28°°C 0 mm 0% 5 mph 28 % 1017 mb 0 mm/h
Today 10:00 pm
Wetter-Symbol
22° | 22°°C 0 mm 0% 8 mph 47 % 1019 mb 0 mm/h
Tomorrow 1:00 am
Wetter-Symbol
18° | 18°°C 0 mm 0% 4 mph 55 % 1018 mb 0 mm/h
Tomorrow 4:00 am
Wetter-Symbol
19° | 19°°C 0 mm 0% 4 mph 65 % 1018 mb 0 mm/h
Tomorrow 7:00 am
Wetter-Symbol
19° | 19°°C 0 mm 0% 6 mph 64 % 1018 mb 0 mm/h
Tomorrow 10:00 am
Wetter-Symbol
24° | 24°°C 0 mm 0% 6 mph 45 % 1017 mb 0 mm/h
Name Preis24H (%)
Bitcoin(BTC)
€101,255.25
6.68%
Ethereum(ETH)
€2,579.67
8.69%
Fesseln(USDT)
€0.85
0.01%
XRP(XRP)
€2.23
6.97%
Solana(SOL)
€140.91
4.43%
USDC(USDC)
€0.85
0.00%
Dogecoin(DOGE)
€0.170131
9.79%
Shiba Inu(SHIB)
€0.000011
7.31%
Pepe(PEPE)
€0.000011
16.09%
Peanut das Eichhörnchen(PNUT)
€0.246894
20.17%
Nach oben scrollen