🔥 Trending on HN

LLM-assisted reverse engineering: speed, notes, and human judgment

3 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
reverse engineering(ree-VERS en-juh-NEER-ing)

Studying a finished program or device to understand how it works.

LLM(el-el-em)

An AI system that can read and work with language and code.

hypothesis testing(hy-PO-thuh-sis TES-ting)

Testing a possible explanation instead of relying on guesses.

What happened

On November 18, 2026, ioma8 published a personal field report about LLM-assisted reverse engineering. The author says they have studied hardware and software for two decades. During the last six months, AI greatly sped up projects that would have taken years by hand, according to the post. Reverse engineering here means studying a finished program or device from the outside. (original post)

The workflow was not simply “give the AI a file and wait.” First, the author identified the binary’s type, platform, and build. Then the agent recorded discoveries in Markdown. It mapped major functions, branches, and imported components. When the first view was not enough, the author changed the approach and added controlled checks. The human watched the agent’s work and redirected it when needed.

The Hacker News submission reached 246 points and 63 comments. Those numbers measure community attention. They do not prove the source’s claims or results. (Hacker News thread)

Background

The post lists DeepSeek Harness with DeepSeek v4.1 Flash, Ghidra with an MCP bridge, and command-line tools such as radare2 and kuna. The author sometimes uses Codex as well. The more important idea is keeping a durable record of the investigation. Notes help the person understand the work. They also help an agent continue after a long session or handoff.

Why it matters

Reverse engineering can help people understand old software, repair devices, or build compatible tools. If an AI can organize a long investigation, it may reduce tedious work for an experienced researcher. Yet the author still chooses what to inspect and when to change direction. The report therefore presents a division of labor. The model handles repeated analysis. The person supplies context, judgment, and course corrections.

What the source establishes

The author’s most distinctive practice is hypothesis testing. First, write down a possible explanation. Next, design a test that can support or reject it. Record the result. Then repeat the cycle. This keeps the agent from guessing forever.

The post also describes Python with uv for quick probes and prototypes. Rust is preferred for finished changes. The author values the compiler’s strict feedback during iteration. Claimed results include reverse engineering the Alphasmart NEO typing device, mapping communications for EmWave and Sony Alpha cameras, building custom firmware for several music devices and the Xteink X4 reader, and fixing an old HP printer driver. These are the author’s reported results, not an independent benchmark.

What remains unknown

The post does not provide a controlled comparison with human-only work. “Years by hand” is an estimate, not a measured baseline. The source also cannot separate the model’s contribution from the author’s two decades of experience. We do not know whether another person, model, or toolset would produce the same results.

The examples also touch areas with different legal and safety boundaries. Reverse engineering may serve repair or compatibility, but changing software or devices can affect ownership, security, and permission. This report describes a capability. It is not a permission slip or a step-by-step security guide.

What to watch next

The next useful evidence would compare human-only, AI-only, and human-guided workflows on the same tasks. It would also test whether durable notes and hypothesis testing work outside this author’s projects. The key question is not only whether AI makes analysis faster. It is who checks the result, and who decides where the work is allowed to go.

Sources: Slow Chronicle; Hacker News.

💬 AI lowers the barrier to reverse engineering, but humans keep the responsibility

The comments suggest that LLMs make reverse engineering more accessible, while accuracy, learning value, and legal boundaries still require human judgment. Reported times and results are commenters’ own accounts, not general performance guarantees.

  • One commenter self-reported using Gemini to examine a CCTV camera’s cloud-connection plugin in about two hours. They isolated the network traffic, used Ghidra on matching DLLs, and had the model write a small script to search for authentication material.
  • Another commenter with almost no reverse-engineering experience self-reported examining abandoned audio-effects and DSP instrument software, getting nearly identical sound, and finding clues for a modern reimplementation.
  • Several comments warned that LLMs can loop through guesses on difficult targets. Black-box work needs hypothesis testing, decisive experiments, and supervision by someone who can judge the results.
  • In one user’s experience, IDA Pro worked better than Ghidra for LLM-assisted reverse engineering. They also suggested instrumenting the target, writing periodic reports, and resetting the model’s context.
  • A strong criticism is that delegating the work to AI may produce results faster while removing the skills and satisfaction gained from doing the process yourself.
  • Reverse engineering can be limited by contracts and platform rules; Apple’s developer agreement was offered as one example.
  • Another view treats AI-assisted reverse engineering as a way for consumers to revive abandoned products or games and push back against companies that neglected them.
  • The opposing view is that unlocking a paid Android app can still look like improper use or piracy, even when someone gives educational or privacy-related reasons.

initial digest at 63 comments (revision 1). We fetched 65 comments and sampled 65 across the thread. These are HN users’ reports, not independently verified facts.

🔥 Trending on HN

AI can inspect old software faster. People still guide the search.

📰 Full story: LLM-assisted reverse engineering: speed, notes, and human judgment

One personal report shows how AI and human experience can work together.

1 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
reverse engineering(ree-VERS en-juh-NEER-ing)

Finding how a finished program or device works.

LLM(el-el-em)

A computer helper that reads text and code.

Hacker News(HACK-er NEWS)

A website where people discuss technology stories.

💡 The gist

  • An AI helper can speed up studying old programs and devices.
  • A skilled person still chooses questions and checks answers.
  • Hacker News attention shows interest, not proof.

The author, ioma8, has studied hardware and software for two decades. The post describes reverse engineering. That means finding how a finished program or device works.

The author uses an LLM, a computer helper that reads text and code. First, the person identifies the target. Then the agent writes notes about its findings. The notes matter because long investigations can lose their place. They help both the person and the agent remember earlier discoveries.

When the agent gets stuck, the person changes the question. The author uses a simple loop. First, write a possible explanation. Next, make a small test. Then record what happened. Repeat when needed. This gives the agent evidence instead of endless guesses.

The author also names several tools. These include DeepSeek, Ghidra, radare2, and kuna. The tools are part of the setup. The larger lesson is the teamwork around them.

The post reports results from six months of work. They include studying an Alphasmart NEO typing device. They include mapping communications for Sony Alpha cameras. They also include custom firmware for music devices. The author says some projects would have taken years manually.

That claim needs care. The post is one person’s experience. It does not show a controlled test. It does not show what another researcher would achieve. The author’s twenty years of experience also matter.

The Hacker News post received 246 points and 63 comments. That means the community noticed it. It does not prove every claim. Some reverse-engineering work also involves security or ownership rules. This summary stays at the reporting level.

Next, look for independent comparisons. Useful tests would compare people alone, AI alone, and people guiding AI.

Sources: Slow Chronicle; Hacker News.

💬 AI can make reverse engineering easier, but checking and rules still matter

Reverse engineering means studying a finished program to work out how it functions. The comments include both success stories and warnings about relying on AI.

  • One commenter said, as a personal report, that Gemini helped inspect camera software in about two hours. AI helped organize the network and program evidence.
  • A person with little reverse-engineering experience said AI helped them understand old sound software and find ideas for recreating it.
  • AI can keep guessing without enough evidence. People must run tests, check the answers, and choose tools such as IDA Pro or Ghidra carefully.
  • Some people say AI saves time; others say doing the work yourself is how you learn and feel the achievement.
  • Contracts and rules may restrict reverse engineering. Some see reviving abandoned software as helping consumers, while others see unlocking a paid app as piracy.

initial digest at 63 comments (revision 1). We fetched 65 comments and sampled 65 across the thread. These are HN users’ reports, not independently verified facts.

🔥 Trending on HN

People and AI look inside old machines

📰 Full story: LLM-assisted reverse engineering: speed, notes, and human judgment

AI helps find clues. A person chooses the next step.

1 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
reverse engineering(ree-VERS en-juh-NEER-ing)

Finding how a thing works.

LLM(el-el-em)

A computer helper that reads words and instructions.

Hacker News(HACK-er NEWS)

A place where people share technology stories.

ioma8 wrote about looking inside old machines and programs. People call this reverse engineering. It means finding how a thing works.

An LLM is a computer helper. It reads words and computer instructions. The helper makes notes about its discoveries. When it gets stuck, the person chooses a small test. Then they look at the result together.

The writer says this helped during six months. The writer studied an old typing machine. The writer studied cameras and music machines too. That is the writer’s own report. We do not know if everyone would work as fast.

Hacker News is a place for technology stories. The post got 246 points and 63 comments. That shows attention. It does not show perfect truth. People must also respect owners’ rules.

Sources: Slow Chronicle; Hacker News.

💬 AI can help look inside software, but people still need to check

Some people say AI made it easier to learn how software works. The result is not guaranteed for everyone.

  • One person said an AI helped them look inside a camera program in about two hours. That is only their personal report.
  • Another said AI helped with an old sound program. But AI can guess wrong, so a person must test it.
  • Doing the work yourself can teach you more and feel better.
  • There are rules. Helping an abandoned product and unlocking a paid app are not judged the same by everyone.

initial digest at 63 comments (revision 1). We fetched 65 comments and sampled 65 across the thread. These are HN users’ reports, not independently verified facts.

Sources