#LLMrisk
LightNews LightNews
toxsec.bsky.social
On the security side, prompt injection is having its “log4j but semantic” moment: new research and vendor writeups are all saying the same thing, agents are trivially hijacked through hostile inputs, jailbreak chains, and indirect injections.
#ai #tech #cyber

genai.owasp.org/llmrisk/llm0...
November 20, 2025 at 5:23 PM
getnews-me.bsky.social
A systematic review of 70 peer‑reviewed studies finds that LLMs can boost efficiency but risk superficial understanding, bias and hallucinations in education. Read more: https://getnews.me/large-language-models-pose-superficial-learning-risks-in-education/ #AIeducation #LLMrisk
Large Language Models Pose Superficial Learning Risks in Education
September 29, 2025 at 10:35 AM
nymphatska.bsky.social
genai.owasp.org/llmrisk/llm0...
LLM04:2025 Data and Model Poisoning
Data poisoning occurs when pre-training, fine-tuning, or embedding data is manipulated to introduce vulnerabilities, backdoors, or biases. This manipulation can compromise model security, performance,...
genai.owasp.org
June 10, 2026 at 11:45 AM
christinaayiotis.bsky.social
“While traditional software vulnerabilities focus on issues like code flaws & dependencies, in ML the risks also extend to third-party pre-trained models & data. These external elements can be manipulated through tampering or poisoning attacks.” genai.owasp.org/llmrisk/llm0... @owasp.org
LLM03:2025 Supply Chain
LLM supply chains are susceptible to various vulnerabilities, which can affect the integrity of training data, models, and deployment platforms. These risks can result in biased outputs, security brea...
genai.owasp.org
July 26, 2025 at 10:28 PM
8-p.info
OWASP にも LLM のページがあるの時代を感じる。
genai.owasp.org/llmrisk/llm0...

しかし "Provide specific instructions about the model’s role, capabilities, and limitations within the system prompt." より "treating the model as an untrusted user" が先じゃないかなあ。システムプロンプトに書いてあること守れたら苦労しないような。
LLM01:2025 Prompt Injection
A Prompt Injection Vulnerability occurs when user prompts alter the LLM’s behavior or output in unintended ways. These inputs can affect the model even if they are imperceptible to humans, therefore p...
genai.owasp.org
August 28, 2025 at 2:24 AM
syrtica.bsky.social
One way: "Indirect prompt injections occur when an LLM accepts input from […] websites or files. The content may have […] data that when interpreted by the model, alters the behavior of the model in unintended or unexpected ways."
Including manipulating the output.
genai.owasp.org/llmrisk/llm0...
LLM01:2025 Prompt Injection
A Prompt Injection Vulnerability occurs when user prompts alter the LLM’s behavior or output in unintended ways. These inputs can affect the model even if they are imperceptible to humans, therefore p...
genai.owasp.org
June 27, 2026 at 10:08 PM
lexthirteenai.bsky.social
A pipeline with no feedback loop is a static system pretending to be intelligent. Drift is not a question of if.#AIinproduction #MLops #ModelDrift #LLMrisk
September 13, 2025 at 8:00 PM
jarekceborski.com
LLM01:2025 Prompt Injection:
genai.owasp.org/llmrisk/llm...
LLM01:2025 Prompt Injection - OWASP Top 10 for LLM & Generative AI Security
A Prompt Injection Vulnerability occurs when user prompts alter the LLM’s behavior or output in unintended ways. These inputs can affect the model even if they are imperceptible to humans, therefore prompt injections do not need to be human-visible/readable, as long as the content is parsed by the model. Prompt Injection vulnerabilities exist in how […]
genai.owasp.org
January 1, 2025 at 9:13 PM
lsaiz.bsky.social
Pregunta para juristas:

¿La inclusión de textos que pretendan "engañar" a una IA en documentos contractuales podría ser considerado como estafa?

Estoy pensando en inyección indirecta genai.owasp.org/llmrisk/llm0...

¿Qué sería aquí "engaño bastante"?
LLM01:2025 Prompt Injection
A Prompt Injection Vulnerability occurs when user prompts alter the LLM’s behavior or output in unintended ways. These inputs can affect the model even if they are imperceptible to humans, therefore p...
genai.owasp.org
May 6, 2026 at 9:08 AM
tk1ng.bsky.social
LLM04:2025 Data and Model Poisoning genai.owasp.org/llmrisk/llm0...

"Backdoor Insertion​
Split-View Data Poisoning​
Frontrunning Poisoning​
Prompt Injection​
Toxic Data Injection​
Supply Chain Poisoning​
Malicious Pickling​
Sleeper Agent Creation​
Misinformation Doc Insertion​"
Bias induction"
LLM04:2025 Data and Model Poisoning
Data poisoning occurs when pre-training, fine-tuning, or embedding data is manipulated to introduce vulnerabilities, backdoors, or biases. This manipulation can compromise model security, performance,...
genai.owasp.org
October 13, 2025 at 2:13 PM