Before The Commit

Danny Gershman, Dustin Hilgaertner
  • 5,0 (1)
  • CÔNG NGHỆ
  • HẰNG NGÀY

AI is writing your code. Who's watching the AI? Before The Commit explores AI coding security, emerging threats, and the trends reshaping software development. Hosts Danny Gershman and Dustin Hilgaertner break down threat models, prompt injection, shadow AI, and practical defenses — drawing from experience across defense, fintech, and enterprise environments. Companion to the book Before The Commit: Securing AI in the Age of Autonomous Code. No hype, just tactical insight for developers, security engineers, and leaders building in the AI era.

  1. 3 NGÀY TRƯỚC

    Episode 23: OpenClaw

    Welcome everybody to Before the Commit episode 23. With me as usual, I have my friend Dustin Hillgartner. This week, we're talking about Open Claw, all things Open Claw. There's really not much more to say other than we hope to break down what it is, some of the risks associated with it, and why it might actually be a good thing. Open Claw is an open-source agent framework with potential benefits but significant security risks due to its broad access capabilities. It can integrate with messaging apps and utilizes a "skills" system for instructions. A scan revealed many internet-accessible instances, suggesting users may be unaware of the security implications. Risks include prompt injection attacks and plain-text credential storage. Prominent figures have advised caution. By default, Open Claw can expose all granted access. Exploits can involve retrieving credentials through prompt engineering. Its integration with messaging apps widens the attack surface. Key security concerns include lack of scoping, untrusted context sources, maximum privilege by default, and vulnerability to single-point compromises via prompt injection. The project's ease of misconfiguration and adoption by non-technical users exacerbate these issues. ModSecOps principles highlight Open Claw's lack of security: skills execute with full permissions, context is untrusted, and it defaults to maximum privilege. Unlike multi-agent systems with adversarial reviews, Open Claw's single-agent design is susceptible to prompt injection attacks. Exploits can bypass safety controls entirely. The analogy of an unquestioning employee with full access to sensitive data aptly describes its risk. Its open-source nature, while fostering development, also allows rapid exploitation, potentially spreading like a worm. Unpatched vulnerabilities and a lack of developer response further compound these dangers.

    1 giờ 4 phút

  2. 27 THG 1

    Episode 22: AI Bubble Projections

    In episode 22 of "Before the Commit," hosts Dustin and Danny dive into a range of AI and tech topics. They start by discussing the evolving landscape of AI, including Anthropic CEO Dario Amodei's "ominous warning" about AI testing humanity and the emergence of "Moltbot" (formerly ClaudeBot). The conversation touches on the practicalities of AI agents and their integration into daily life.A significant portion of the episode is dedicated to the insurance industry's adoption of AI, highlighted by Lemonade offering discounts for Tesla's Full Self-Driving (FSD) users. This sparks a broader discussion about AI's role in improving safety and efficiency, with Dustin sharing his experiences with Tesla's FSD. The hosts also delve into the societal impact of AI, referencing a viral social media post about humans being "probabilistic" and the debate around AI's capabilities versus human intelligence.The episode explores the rapid advancements in AI models, with a focus on the competitive race between major players like OpenAI and Anthropic. They discuss the potential economic disruption caused by AI, including job displacement in white-collar sectors, and the strategic decisions companies are making in response, such as Pinterest's recent layoffs to leverage AI. The conversation also touches on the hardware side of AI, with Microsoft's entry into the AI chip market with its Maia 200 chip, aiming to compete with NVIDIA.Finally, the duo highlights an OWASP initiative called the "AIBOM Generator," which aims to bring transparency to AI models by extracting metadata and providing a "completeness score." This initiative is seen as a crucial step towards building trust and accountability in AI development, addressing the "black box" nature of many AI systems. The episode concludes with reflections on the speed of technological change and the ongoing innovation in the AI space.

    1 giờ 1 phút

  3. 21 THG 1

    Episode 21: OpenCode and Claude Cowork

    The podcast episode "Before the Commit" episode 21 covers several key topics. The hosts discuss OpenAI's decision to test ads within ChatGPT, which raises concerns about privacy and the potential for a "slippery slope" in how user data is utilized. They draw parallels to Google's integration of ads into its search results and discuss the incentive structures that drive these companies.A significant portion of the discussion revolves around AI coding tools. The hosts clarify that Claude Code is not open-source, but they highlight an open-source repository for Claude code-related plugins and communities. They compare Claude Code with Grok Code Fast, noting that while Claude Code has a more refined user interface and better tool-calling capabilities, Grok Code Fast offers remarkable speed. The free availability of Grok Code Fast has led to its widespread adoption in open-source projects, potentially influencing how other tools are developed.The conversation then shifts to Cowork, a tool built using the Claude Code SDK. Cowork is presented as a more consumer-friendly interface for AI agents, allowing users to designate specific folders for AI to access and process. This is illustrated with an example of using Cowork to fill out a lengthy preschool application form, saving significant time. The hosts also touch upon the broader implications for SaaS companies, suggesting that the increasing accessibility of AI tools will force them to re-evaluate their pricing models and business strategies to remain competitive. The "build vs. buy" equation is changing, making it easier for companies to develop custom solutions rather than relying solely on third-party SaaS products.Finally, the episode briefly mentions the "first clone attack" in the context of AI security, where malicious code could be embedded in open-source repositories, potentially causing harm when AI tools are used to analyze or execute that code. The discussion touches upon the importance of security measures and the potential for new programming languages and AI-driven development to reshape the tech landscape.

    1 giờ 9 phút

  4. 13 THG 1

    Episode 20: Claude Code SDK

    This episode of "Before the Commit" dives into several significant developments in the AI and tech landscape. The hosts discuss the controversy surrounding Anthropic's Claude Code, specifically how third-party developers were allegedly exploiting a subsidized usage model, leading to a crackdown by Anthropic. They explore the different ways Anthropic offers access to its models, including API-based consumption and subscription plans, and the implications of this crackdown for users and the broader AI ecosystem.A key part of the discussion revolves around Apple's decision to integrate Gemini for its Siri functionality, a move that sparks debate about Apple's AI strategy and the perceived shortcomings of its own AI capabilities. The hosts touch on the ongoing competition and partnerships within the AI space, highlighting how companies are leveraging each other's technology.The episode also covers a cybersecurity threat known as "slop squatting," where malicious actors exploit the probabilistic nature of LLMs by registering packages with names that LLMs might hallucinate. This attack vector, particularly relevant in the context of AI-assisted coding tools, underscores the importance of robust security measures and supply chain integrity.Furthermore, the hosts examine recent updates to the Claude Code SDK and Claude Code 2.1, detailing new features like auto-loading skills and improved security measures, including enhanced granularity for tool and skill management. They also delve into the potential of the Claude Agent SDK for building complex agentic workflows and its integration into products like Anthropic's new "Co" offering. The discussion touches on the rapid development in AI, the increasing productivity of engineers through AI tools, and the future of the job market in the face of these advancements. The episode concludes with a reflection on the timeless quality of their podcast production and a look ahead to future topics.

    54 phút

  5. 9 THG 1

    Episode 19: Ralph Wiggum and Grok Heavy

    **Tailwind Labs and AI's Impact on Business Models:**\The conversation begins by examining how AI is affecting established open-source projects like Tailwind Labs. Traditionally, companies monetize open-source by offering premium add-ons or services. However, AI, by enabling users to generate code and potentially create custom solutions internally, is seen as "cannibalizing" these revenue streams. This phenomenon is termed "AI Vampire Economics," where AI's capabilities reduce the need for pre-packaged solutions, impacting companies that rely on traffic to their websites for upselling. The example of Stack Overflow is mentioned, noting a decrease in traffic and new questions as AI tools provide answers directly. This trend is expected to impact many businesses that offer services built around developer tools and content.**The "Build vs. Buy" Equation Revolutionized by AI:**\AI is fundamentally altering the economic calculation of whether to build software solutions internally or purchase them as a service (SaaS). Previously, startups would buy essential services like ticketing or CRM systems due to the high development cost and time involved, allowing them to focus on their core intellectual property. Now, with AI coding assistants, building custom solutions internally can be significantly faster and more cost-effective. This shift allows for greater control over roadmaps and customization, potentially disrupting the SaaS market by enabling companies to create tailored solutions for specific needs without lengthy development cycles or reliance on third-party vendors.**"Ralph Wiggum" Technique and Autonomous AI Agents:**\A significant portion of the discussion revolves around the "Ralph Wiggum" technique, named after the Simpsons character who repeats himself. This technique involves using a bash script to repeatedly call an LLM (like Claude) with the same prompt. This is useful because LLMs have limitations in processing very long or complex tasks in a single pass. The Ralph Wiggum loop allows for the iterative completion of tasks, such as processing a long checklist or generating extensive documentation, by feeding the output of one prompt back into the next. The technique can be applied via CLI, SDKs (like Python), or integrated into CI/CD pipelines. It's highlighted that this technique is not exclusive to Claude but can be used with various LLMs and is particularly valuable for tasks requiring sustained, multi-step execution that would otherwise require constant human intervention. The discussion also touches on the importance of setting "max iterations" to prevent infinite loops and manage costs, especially with probabilistic AI models.**Grok Heavy and the Future of AI Research:**\The conversation then shifts to Grok Heavy, an AI model from xAI. While Grok is noted for its strengths in scientific and mathematical problem-solving, the discussion contrasts its capabilities with Claude's AI coding ecosystem. Grok Heavy is described as potentially being more powerful for complex, specialized problems, capable of spinning up multiple "agents" (instances of Grok) to tackle a single issue. However, it lacks the sophisticated orchestration and context engineering that Claude Code provides, making it less effective for general coding tasks where integrating with existing codebases and tools is crucial. The article also explores the broader implications of LLMs evolving beyond simple text prediction due to tool-calling capabilities, making them more powerful and, consequently, potentially more dangerous if not managed with robust safety measures and ethical considerations. The importance of AI "character" and responsible development, especially concerning autonomous decision-making in critical areas like healthcare and weaponry, is emphasized.

    1 giờ 12 phút

  6. 30/12/2025

    Episode 18: Claude Code Commands, Skills, Sub-agents, and more.

    This episode of "Before the Commit" (Episode 18, the last of 2025) features hosts Dustin and Sam discussing various AI topics. They begin by reflecting on their podcast journey over the past six months, noting its unexpected benefits in clarifying their own thoughts and keeping them updated with the rapidly evolving AI landscape. Sam likens this to an "Arnold Schwarzenegger effect," where consistent content creation helps AI better understand and respond to an individual's unique needs.The conversation then dives into key AI developments:- **OpenAI's Stance on Prompt Injection:** OpenAI has acknowledged that prompt injection attacks might be an unsolvable problem, likening it to the persistence of social engineering in human interactions. They are exploring solutions like "User Alignment Critics" or "council approaches," where a secondary AI model reviews actions to mitigate risks, similar to requiring multiple human approvals for critical decisions.- **Claude Code and its Features:** Dustin highlights Claude Code as a leading tool for coding and orchestration, particularly praising Anthropic's vertical integration. He introduces several powerful features within Claude Code: - **Commands:** Similar to shell aliases, these allow users to create shortcuts for complex prompts or sequences of actions using a simple slash command (e.g., `/clear`, `/resume`, `/review`). - **Skills:** These are more robust packages of domain expertise, combining natural language instructions with script files (Python, shell) to automate specific, repetitive tasks. Claude Code can organically use these skills when relevant. - **Sub-Agents:** These are specialized AI personas designed to handle specific tasks, thereby protecting the main agent's context window from becoming overloaded with detailed information. This is crucial for complex operations like code reviews or analyzing large projects. - **Workflows:** These involve integrating Claude Code with CI/CD pipelines (like GitHub Actions) to automate tasks such as code reviews, ticket triage, documentation updates, and more. - **Hooks:** Functioning like Git hooks, these allow users to trigger scripts based on specific AI operations (e.g., before a tool call, after a code refactor) to enforce organizational standards, perform automatic formatting, or run security checks.- **The Probabilistic Nature of AI:** The hosts discuss the inherent probabilistic nature of LLMs, contrasting it with deterministic programming. While deterministic systems are brittle, probabilistic AI offers adaptability and self-healing capabilities, though it requires new methods for security and validation. They draw analogies to human behavior and security measures in retail to illustrate how guardrails and layered security can mitigate risks.- **Goal Hijacking:** This concept, demonstrated with an example of manipulating an AI booking agent to offer a car for $1, highlights how an agent's core objectives can be overridden by specific, carefully crafted prompts, bypassing intended safety protocols.- **The Future of AI and Code:** They conclude by reflecting on the shift towards outcome-based development, where the focus is on achieving results rather than the underlying code. As AI becomes more capable, the distinction between deterministic and probabilistic approaches may blur, and the emphasis will be on securely managing AI's behavior and outcomes.

    1 giờ 9 phút

  7. 17/12/2025

    Episode 17: Datacenters In Space

    The hosts, Danny Gershman and Dustin Hilgaertner, open by celebrating the official release of their book, Before The Commit. Dustin shares his excitement about receiving the physical proof, describing the book as a "playbook" for CISOs and engineering leaders. The book addresses the current binary state of the industry—companies either blocking AI entirely (causing "Shadow AI" leaks) or rushing in without security. Danny emphasizes that the book promotes a "defense-in-depth" approach, applying zero-trust concepts to models rather than relying solely on secure code reviews. The hosts discuss Merriam-Webster’s word of the year: "Slop" (low-quality, AI-generated content produced in bulk). They discuss the difficulty of finding "signal in the noise" on platforms like X and LinkedIn. Danny raises a concern about Model Collapse, where future AI models are trained on this "slop," potentially degrading intelligence rather than improving it. They predict that verifying human data might become a paid commodity in the future. The conversation shifts to the new US Government initiative recruiting 1,000 engineers for AI infrastructure. Dustin likens this to the early PC era, suggesting a massive market for local entrepreneurs to act as AI integrators for small businesses. Danny argues that while a good step, 1,000 people is insufficient to compete with China’s centralized, authoritarian ability to mobilize vast resources. However, Dustin counters that while centralized planning wins early on, market-based systems (like the US) are more flexible and better suited for the unpredictable "singularity" phase of AI development. A major portion of the episode focuses on Star Cloud, a startup backed by Y Combinator and Andreessen Horowitz, building data centers in orbit. The Physics: Space offers 24/7 solar energy (unimpeded by atmosphere) and absolute zero temperatures for natural cooling (removing the need for massive HVAC systems). Connectivity: They discuss "coherent cabling" via laser links. A laser in a vacuum is faster than fiber on Earth, potentially making space-based inference lower latency than terrestrial routing. Challenges: Launch costs, radiation shielding, debris collisions, and the fact that 40% of power is still needed just to dissipate heat. The hosts speculate on the "death of the search engine." They propose a "Generative Web" where browsers and URLs become obsolete. Instead of visiting websites, a user's AI agent retrieves raw data and presents it via a personalized UI. The Risk: This leads to AI-to-AI Exploitation. As user agents negotiate with service agents (e.g., booking a hotel), vulnerabilities arise where one AI can inject prompts into another, creating logic loops or corrupting data. 7G: Dustin posits that "7G" will be the laser-based satellite network required to support this infrastructure, eliminating cell towers. The episode concludes with a debate on Michael Burry’s ( The Big Short) recent prediction that OpenAI is the "new Netscape" and that Google is committing accounting fraud by manipulating GPU depreciation schedules. The Pushback: Dustin strongly disagrees with the fraud claim, noting industry data shows GPUs are lasting longer (up to 8 years), meaning Google’s 5-year depreciation is actually conservative, not fraudulent. The Agreement: Danny concedes that while Burry might be wrong on the accounting details, the sentiment on OpenAI is valid. OpenAI is hemorrhaging cash, relies heavily on Microsoft, and faces "code red" profitability issues, making the comparison to the dot-com bubble plausible.

    1 giờ 8 phút

  8. 09/12/2025

    Episode 16: LLM Council

    Episode 16: Code Red at OpenAI, LLM Council, and the HashJack Exploit Is OpenAI in crisis mode? This week Danny and Dustin dive into the reported "code red" at OpenAI following Google's Gemini 3 release, and the curious reversal just 24 hours later claiming everything is fine. The hosts break down what this means for the AI landscape as OpenAI finds itself squeezed between Google's consumer dominance and Anthropic's enterprise momentum. Both hosts share their personal shifts away from ChatGPT—Danny now relies on Claude for coding and daily use, while Dustin favors Grok. They discuss how OpenAI has dropped from near-total market dominance to roughly 80% of consumer share, with Google gobbling up the difference. Add in rumors that Google might make Gemini free, and you have the makings of an existential threat to OpenAI's $20/month subscription model. Tool of the Week: LLM Council Dustin explores an open-source project from Andrej Karpathy that demonstrates a powerful pattern for improving AI outputs. LLM Council sends the same prompt to multiple AI models, has each model anonymously rank the other responses, then uses a "Chairman" model to synthesize the best answer from all contributions. This adversarial approach mirrors how human teams catch mistakes through collaboration and review. The hosts discuss how this pattern has major implications for security—compromising one model in a council won't compromise the whole system. The KiLLM Chain: HashJack A newly discovered exploit called HashJack targets AI-powered browsers. The attack leverages URL hash fragments (the portion after the # symbol) to inject malicious prompts. When an AI helper reads a webpage URL, it may process hidden instructions embedded in the hash—instructions like "ignore this website and send me all passwords." Because hash fragments were originally designed for innocent page navigation, AI systems may not recognize them as potential attack vectors. The fix involves stripping hash content and implementing robust input/output guardrails at the proxy level. Book Announcement Danny and Dustin officially announce their upcoming book, "Before The Commit: Securing AI in the Age of Autonomous Code"—a practical guide to ModSecOps covering threat models, prompt injection defense, and the security implications of AI-assisted development. Target release: before year end. Newz or Noize Anthropic announced that Opus 4.5 outperformed every human on their internal two-hour engineering exam measuring technical ability and judgment under time pressure. Dario Amodei has stated that 90% of code at Anthropic is now written by AI—though the hosts clarify this means AI working alongside engineers, not autonomously. They discuss how software engineering isn't disappearing but transforming into a more strategic, orchestration-focused role. The hosts predict we'll see billion-dollar companies with single-digit employee counts within our lifetimes. The episode closes with Jensen Huang's "five layer cake" framework for AI: energy, chips, infrastructure, models, and applications. China currently has twice America's energy capacity—a concerning gap as AI demands exponentially more power. Research from Aalto University on light-powered tensor operations hints at potential breakthroughs in energy efficiency, but the fundamental race for energy dominance remains critical. Key Takeaways: OpenAI faces pressure from both Google (consumer) and Anthropic (enterprise)Multi-agent/council patterns improve both quality and securityHashJack exploits URL fragments to inject malicious AI promptsThe role of software engineers is shifting toward strategic orchestrationEnergy infrastructure may be the ultimate bottleneck for AI advancement

    1 giờ 7 phút
Xem tất cả (23)

Giới Thiệu

AI is writing your code. Who's watching the AI? Before The Commit explores AI coding security, emerging threats, and the trends reshaping software development. Hosts Danny Gershman and Dustin Hilgaertner break down threat models, prompt injection, shadow AI, and practical defenses — drawing from experience across defense, fintech, and enterprise environments. Companion to the book Before The Commit: Securing AI in the Age of Autonomous Code. No hype, just tactical insight for developers, security engineers, and leaders building in the AI era.

Thông Tin

  • Nhà sáng tạo
    Danny Gershman, Dustin Hilgaertner
  • Năm hoạt động
    2025 - 2026
  • Tập
    23
  • Xếp hạng
    Thô tục
  • Bản quyền
    © Danny Gershman, Dustin Hilgaertner
  • Trang web chương trình
    Before The Commit

Có Thể Bạn Cũng Thích