Developer builds AI agent security system PHAGE for Google Cloud Hackathon

·1 views

A developer spent August building PHAGE, an AI security platform designed to protect fleets of AI agents from prompt-injection and tool-poisoning attacks. The system continuously tests deployed agents with tailored attack payloads, quarantines those that fail, and stores attack signatures to neutralize repeat threats automatically. PHAGE was created for organizations running AI agents without dedicated security staff. The project was submitted to the Google Cloud All Things Agentic Hackathon and built entirely within the submission window with permitted AI assistance. A live, read-only dashboard is publicly accessible, displaying evaluation artifacts and a redacted span database without requiring a Google account.

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)

Developer consolidates nine brands onto one framework, adds payments in a week

A developer at ITW Creative Works spent the week consolidating multiple products under a unified internal framework called OMEGA, releasing ten incremental updates over seven days. Key additions included end-to-end payment support, a single analytics contract, and shared lifecycle features such as retention dialogs and offline mode, all inheritable by every brand. Nine products — including StudyMonkey, Giftly, DashQR, and others — were migrated from legacy setups onto the framework within a single hour using a shared brand template and config conversion. A new video pipeline project called 'cutroom' was also started, aiming to automatically turn long recording sessions into edited videos without manual timeline work. The developer plans to ship the first video output and continue the brand rollout in the coming week.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Four Java Bugs That Misleading Indentation Can Hide From Developers

In Java, whitespace has no meaning to the compiler, which processes code purely as tokens — but humans rely on indentation to understand logic, creating a dangerous gap. Common bugs hidden by poor indentation include a misplaced semicolon that turns a conditional into a no-op, a dangling else that attaches to the wrong if statement, and a closing brace that ends a block one line too early after a refactor. A fourth pattern, the unbraced guard chain, mirrors the infamous CVE-2014-1266 Apple 'goto fail' bug, where a duplicated line bypassed SSL certificate verification on iOS for over a year. Static analysis tools like SonarQube can catch some of these patterns, but only if they are actually run. Developers are advised to always brace single-statement blocks, enable format-on-save in their IDE, and configure build pipelines to fail on formatting drift.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

CISA: Over 100 Water Sector OT Systems Targeted in July 2026 Cyberattacks

The U.S. Cybersecurity and Infrastructure Security Agency reported that more than 100 internet-exposed systems in the water and wastewater sector were targeted by malicious actors in July 2026. Attackers identified and targeted programmable logic controllers connected directly to cellular modems, exploiting configurations that allowed public internet reachability without passing through corporate IT networks. Iran-linked threat actors have been cited in broader attribution related to this activity. No critical operational disruptions, water outages, or water quality impacts have been confirmed as a result of the attacks. CISA has urged water utilities to remove direct internet exposure, enforce strong credentials, apply firmware updates, and implement multi-factor authentication to reduce attack surface.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Build Timezone-Aware Reminder Queues Without Silent Delivery Failures

A technical guide published on DEV Community outlines a robust approach to scheduling user reminders in Node.js for services spanning multiple time zones, such as US and EU edtech platforms. The core recommendation is to store schedules using IANA timezone identifiers and local clock times, while persisting the next due instant in UTC for reliable querying. The article warns against naively adding fixed intervals to previous UTC timestamps, which breaks during Daylight Saving Time transitions that can skip or duplicate local times. To prevent delivery backlogs, the guide proposes a state model using durable per-reminder cursors, unique occurrence IDs, and bounded worker claim sizes to keep lateness visible and auditable. Code examples in Go illustrate the calendar boundary logic, with the author noting the same contract applies to Node.js using any compatible time library.

0 comments Read more at DEV Community