Tom Bonner (@thomas_bonner) header
Tom Bonner (@thomas_bonner) profile picture

Tom Bonner

@thomas_bonner

SVP of Research @hiddenlayersec. Formerly Norman, HP, Cylance, BlackBerry. All views are my own.

 England, United Kingdom
hiddenlayer.com/
Joined December 2009
  • Tweets510
  • Following1.1K
  • Followers1.3K
64Media
Photo by Tom Bonner (@thomas_bonner)Photo by Tom Bonner (@thomas_bonner)Photo by Tom Bonner (@thomas_bonner)Photo by Tom Bonner (@thomas_bonner)Photo by Tom Bonner (@thomas_bonner)Photo by Tom Bonner (@thomas_bonner)Photo by Tom Bonner (@thomas_bonner)Photo by Tom Bonner (@thomas_bonner)
During the July 2026 Hugging Face intrusion, an autonomous AI agent stashed its own working files in public repos. We found 500+ of them - tool source, C2 code, exfil'd data, etc. - and analyzed the whole lot. #AI #CyberSecurity #AISecurity #ThreatIntel hiddenlayer.com/research/hug…
Hugging Face AI Agent Attack: What the Artifacts Reveal

Hugging Face AI Agent Attack: What the Artifacts Reveal

HiddenLayer analyzes artifacts from the Hugging Face AI agent incident, revealing how an autonomous agent staged tools, exfiltrated data, and operated in the wild.

hiddenlayer.com
1
4
1
11
3,301
CVE-2026-45829: ChromaToast Served Pre-Auth Demo Demonstration of CVE-2026-45829 from the "ChromaToast Served Pre-Auth" blog.
8
5
34
7,725
Max-severity flaw in ChromaDB for AI apps allows server hijacking bleepingcomputer.com/news/se… bleepingcomputer.com/news/se…
Max-severity flaw in ChromaDB for AI apps allows server hijacking

Max-severity flaw in ChromaDB for AI apps allows server hijacking

A max-severity vulnerability in the latest Python FastAPI version of the ChromaDB project allows unauthenticated attackers to run arbitrary code on exposed servers.

bleepingcomputer.com
1
8
3
31
5,005
Fake OpenAI repository on Hugging Face pushes infostealer malware bleepingcomputer.com/news/se… bleepingcomputer.com/news/se…
Fake OpenAI repository on Hugging Face pushes infostealer malware

Fake OpenAI repository on Hugging Face pushes infostealer malware

A malicious Hugging Face repository that reached the platform's trending list impersonated OpenAI's "Privacy Filter" project to deliver information-stealing malware to Windows users.

bleepingcomputer.com
1
21
2
60
6,936
Tom Bonner retweeted
Jim Simpson (@4a696d) profile picture
Jim Simpson@4a696d17 Jun 2025
So this one has been a while coming. If you've seen any of the talks or workshops I've done recently you'll have heard me talking about the way we are thinking about attacks on generative AI. Well, it's finally here. hiddenlayerai.github.io/ape-…
Image from Jim Simpson (@4a696d)
5
1
16
2,251
Announcing our latest attack technique, "Policy Puppetry" - a single, transferable prompt blending structured policy & roleplay that bypasses alignment in frontier AI models. Game-changing for red-teaming! #AI #GenAI #RedTeam #CyberSecurity hiddenlayer.com/innovation-h…
Universal AI Bypass: How Policy Puppetry Leaks System Prompts and Safety Data

Universal AI Bypass: How Policy Puppetry Leaks System Prompts and Safety Data

HiddenLayer’s latest research uncovers a universal prompt injection bypass impacting GPT-4, Claude, Gemini, and more, exposing major LLM security gaps.

hiddenlayer.com
1
1
6
418
HiddenLayer researchers uncovered a malicious version of the Android #DeepSeek - #AI Assistant app recently uploaded to a popular #malware scanning service.
Image from HiddenLayer (@hiddenlayersec)
4
2
15
1,189
Our latest research, modifying the computational graphs of neural networks to introduce backdoors, has been extremely interesting. In this blog, we've targeted ResNet, YOLO, and Phi-3 models, using different triggers and payloads. #AI #CyberSecurity hiddenlayer.com/research/sha…
ShadowLogic: Persistent No-Code Backdoors in AI Computational Graphs

ShadowLogic: Persistent No-Code Backdoors in AI Computational Graphs

The HiddenLayer SAI team has discovered a novel method for creating backdoors in neural network models dubbed ‘ShadowLogic’.

hiddenlayer.com
2
2
517
ShadowLogic Attack Targets AI Model Graphs to Create Codeless Backdoors securityweek.com/shadowlogic…
ShadowLogic Attack Targets AI Model Graphs to Create Codeless Backdoors

ShadowLogic Attack Targets AI Model Graphs to Create Codeless Backdoors

HiddenLayer details ShadowLogic, a new method of creating codeless backdoors in AI models by manipulating their graphs.

securityweek.com
3
5
1,767
Today, we publicly disclosed 14 new CVEs across three critical machine learning projects. These vulnerabilities highlight the importance of proactive security in AI and reinforce our commitment to safeguarding the future of machine learning. 👉 hiddenlayer.com/sai-security…
Image from HiddenLayer (@hiddenlayersec)
2
5
337
🚨 New Research: AI’ll Be Watching You Our team recently presented this cutting-edge research at @aivillage_dc 2024. This blog takes a deep dive into the security of edge AI devices, starting with the widely used Wyze Cam 👉 hubs.ly/Q02KWPQ70 #EdgeAI #AIResearch
Image from HiddenLayer (@hiddenlayersec)
1
1
8
256
Tom Bonner retweeted
Costin Raiu (@craiu) profile picture
Costin Raiu
@craiu22 May 2024
Small thread about my experience YARA-X, version 0.3.0 fresh beta, courtesy of @plusvic
3
37
3
113
26,583
Tom Bonner retweeted
Abraxus (@Abraxus7331) profile picture
Abraxus@Abraxus73312 May 2024
I'm incredibly excited to announce that I will be speaking with @thomas_bonner this August at #BHUSA. If you enjoyed the research on R this week come check out our talk to see even more! blackhat.com/us-24/briefings…
1
1
8
1,525
👨‍💻🔐 A new security #vulnerability (CVE-2024-27322) has been discovered in the R #programming language. It could allow attackers to execute arbitrary code through malicious RDS files, exposing your projects to supply chain attacks. Read: thehackernews.com/2024/04/ne… #cybersecurity
New R Programming Vulnerability Exposes Projects to Supply Chain Attacks

New R Programming Vulnerability Exposes Projects to Supply Chain Attacks

A critical vulnerability (CVE-2024-27322) has been discovered in the R programming language. It could allow attackers to execute arbitrary code.

thehackernews.com
1
75
6
125
25,677
Very nice work from @Abraxus7331 and @KieranEvans89 in discovering CVE-2024-27322, a vulnerability in R's deserialization library that can lead to "R-bitrary" code execution when deserializing untrusted data. hiddenlayer.com/research/r-b…
HiddenLayer Research | R-bitrary Code Execution

HiddenLayer Research | R-bitrary Code Execution

Learn how arbitrary code execution vulnerabilities in AI systems expose models to exploitation and infrastructure compromise risks.

hiddenlayer.com
1
6
1
14
2,641
🎉 We are proud to present the first-ever HiddenLayer AI Threat Landscape Report! Get your copy TODAY to see our survey insights, risks related to the use of #AI & security recommendations for the next 12 months. Learn more 👉 hubs.ly/Q02nmggk0
1
4
8
1,389
Fickling gets powerful new upgrades: a modular analysis API and PyTorch and Polyglot modules. Enhance threat detection and analysis across your ML workflows. buff.ly/3ImyxUI
Image from Trail of Bits (@trailofbits)
1
11
2
40
9,429
🤖 Security researchers have uncovered a new #vulnerability in Hugging Face's Safetensors conversion service that could lead to supply chain attacks, compromising user-submitted models. Read details: thehackernews.com/2024/02/ne… #cybersecurity #hacking #technews
New Hugging Face Vulnerability Exposes AI Models to Supply Chain Attacks

New Hugging Face Vulnerability Exposes AI Models to Supply Chain Attacks

Hugging Face vulnerability allows attackers to hijack machine learning models.

thehackernews.com
1
31
3
61
14,961
Our researchers discovered that the Hugging Face PyTorch to Safetensors conversion service could easily be compromised by attackers, who could tamper with models and leak the token used to create pull requests from the official bot. hiddenlayer.com/research/sil…
Silent Sabotage: Hijacking Safetensors Conversion on Hugging Face

Silent Sabotage: Hijacking Safetensors Conversion on Hugging Face

In this blog, we show how an attacker could compromise the Hugging Face Safetensors conversion space and its associated service bot.

hiddenlayer.com
10
2
17
6,350
📅 SAVE THE DATE: HiddenLayer’s 2024 AI Threat Landscape Report will be released on March 6th. Sign up to be the first to preview the report & join us in a webinar discussion as we share some of the report’s most important findings 👉 hubs.ly/Q02kGr2Q0 #Security4AI
5
6
495
Load more