Why 'The AI Hallucinated' Will Never Fly as an Excuse in a Boardroom

·4 views

A software engineering perspective argues that AI tools like GitHub Copilot cannot be held accountable within corporate structures, creating what the author calls a 'Blame Deficit.' When AI-generated code causes outages or revenue loss, organizations still require a human to own the failure, as accountability chains depend on individuals with professional reputations at stake. The argument holds that this is not a temporary technical limitation but a permanent structural reality of how trust operates in businesses. Rather than fearing AI's ability to generate code quickly, engineers are advised to focus on operational ownership of critical systems as their core professional value. The author promotes these ideas through a recently published book offering frameworks for engineers navigating AI-driven workplaces.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

MCP Tool Descriptions Can Be Silently Altered After Approval, Exposing AI Agents

A security vulnerability in the Model Context Protocol (MCP) ecosystem allows tool descriptions fetched by AI agents to be modified after a user has reviewed and approved them, with no diff, lockfile, or signature to flag the change. Because tool descriptions are loaded directly into an agent's context window on every connection, a malicious or compromised npm package maintainer can inject prompt-injection instructions — such as silently exfiltrating SSH private keys — without touching the user's config file. Existing security scanners like Dependabot, Snyk, and CodeQL do not inspect MCP tool descriptions, leaving the attack surface entirely unmonitored. A tool called Bulwark has been developed to address this gap by generating a content-hash lockfile of all tool description strings, enabling verification of any material changes between sessions. The broader issue stems from four compounding ecosystem properties: tool descriptions acting as prompt content, no pinning mechanism, widespread use of auto-resolving unpinned package fetches, and a lack of organizational inventory of deployed MCP servers.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How CVE Bots and Open Threat Datasets Are Reshaping SecDevOps Pipelines

Modern software development cycles are outpacing traditional security operations, making manual vulnerability scanning a critical liability rather than a reliable safeguard. The industry is shifting toward a 'Security as Code' model where threat intelligence functions as a continuous, automated feedback loop embedded directly into development workflows. Automated CVE tracking bots aggregate data from multiple sources — including NVD API 2.0, OSV.dev, and GitHub Security Advisories — to overcome the latency and rate-limit constraints of relying on any single database. Open threat intelligence datasets, such as blocklists and indicators of compromise from sources like AbuseIPDB and CERT/CC, complement CVE data by identifying actively exploited malicious infrastructure. Building a production-grade system requires an event-driven microservice architecture with a message queue layer to decouple data ingestion from processing, along with a unified internal schema to normalize records from disparate sources.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Linux Landlock LSM lets any user sandbox apps without root privileges

Landlock is a Mandatory Access Control (MAC) security feature built into the Linux kernel since version 5.13. Unlike traditional tools such as SELinux or AppArmor, it allows any user — without administrator rights — to create sandboxes that restrict a process's access to specific file system paths. Once a Landlock restriction is activated, the kernel blocks the sandboxed process from accessing any paths not explicitly permitted, regardless of the user's ID. This makes it significantly easier to limit the damage from exploits, such as a compromised browser, without requiring complex policy languages or root access. The feature integrates directly into the kernel as an additional security layer on top of Linux's existing Discretionary Access Control system.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer fixes AI agent that executed questions and negations as commands

A developer building an interactive AI-driven piece discovered a critical bug where their language parser was triggering state-changing actions based on substring matches alone, without verifying whether a sentence was a question, negation, or quotation. For example, a phrase meaning 'Shall we decide together?' — a question — was incorrectly executed as a 'join' command because it contained a recognized fragment. The root cause was that the system confused the presence of certain characters with an affirmative speech act, effectively treating character detection as authorization. The fix replaced fuzzy substring matching with a strict closed map of pre-approved, exact clauses, each tied to one permitted operation. Any response containing an unregistered sentence now has all its operations disabled, preventing quoted or negated text from being laundered into executable commands.

0 comments Read more at DEV Community