Posts by Albert Ziegler on the XBOW blog
13 posts- AI Research
Grok 4.7 Is Different. So Is the Best Way to Use It.
XBOW had early access to Grok 4.7 and tested it across our offensive security benchmarks. The results were mixed until we split them by orchestration: inside xAI’s Build, Grok 4.7 improved substantially over 4.6.
- AI Research
Engineering the Impossible: Adding Safety to Autonomous Agents
In our Engineering the Impossible blog series, we highlight the thorny offensive security challenges our AI and security experts are tackling, and the genius ways they are overcoming them.
- AI Research
Grok 4.5 Is Powerful. The System Around It Makes It Safe.
Grok 4.5 combines powerful offensive-security reasoning with a strong response to XBOW’s independent safety controls, enabling aggressive exploration within controlled boundaries.
- AI Research
Grok 4.5 Poised To Take Over the Middle of the AI Security Market
Grok 4.5 delivers near frontier offensive security performance at a practical price, making it the strongest AI model for many real-world cybersecurity workloads in the middle of the cost curve.
- AI Research
The Rise of Affordable Models: Comparing GLM and Muse Spark on Cyber
Lower-cost AI models like Muse Spark 1.1 and GLM are rapidly improving, making AI-powered offensive security more accessible and changing the economics of vulnerability discovery.
- Security Research
Mythos for Offensive Security: XBOW's Evaluation
We received early access to Mythos Preview for early capability testing a few weeks back. Today, we can finally share what we found.
- AI Research
GPT-5.5: Mythos-Like Hacking, Open to All
We had early access over the past few weeks and tested it across our benchmarks and workflows. Here’s how 5.5 performed for our offensive security capabilities.
- AI Research
Smaller Bites, Bigger Meals: What We Learned Running Opus 4.7 in Offensive Workflows
We got exclusive early access to Anthropic's latest model Opus 4.7. Here's what's new, what's improved, and why it matters for the future of AI security.
- AI Research
AI for Pentesting: Strengths, Weaknesses, and Where XBOW Fills the Gaps
AI is transforming pentesting, but models alone fall short. Learn where AI excels, where it needs structure, and how validation and orchestration make it reliable.
- AI Research
We Ran 1,060 Autonomous Attacks. Here's What the Industry Gets Wrong.
The 2026 AI Safety Report says autonomous attacks aren’t possible yet, but that view changes when considering purpose-built autonomous offensive security systems.
- AI Research
Why LLMs Hallucinate Vulnerabilities
LLMs can suggest vulnerabilities, but confidence isn’t proof. Findings require external validation to distinguish real issues from false positives.
- AI Research
XBOW Unleashes GPT-5’s Hidden Hacking Power, Doubling Performance
OpenAI's initial assessment of GPT-5 showed modest cyber capabilities, yet integrating it into the XBOW platform unleashed its hidden hacking power and doubled our agent's performance
- AI Research
Agents Built From Alloys
This spring, we introduced a novel approach that boosted XBOW’s vulnerability detection performance, increasing success rates from 25% to 40%, and then to 55% on fixed benchmarks.