🔥 Trending on HN

AI Found a Dangerous Crack in Old Software. People Still Had to Fix It.

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

Words
PHP2Shell(pee-aych-pee two shell)

The name of a serious security flaw found in PHP.

technical debt

Old software work that keeps building up and becomes harder to fix.

formal verification

A careful method for checking whether software meets set conditions.

What happened

An IEEE Spectrum report describes PHP2Shell, a serious security flaw in PHP. PHP is a programming language used by many websites. The flaw could expose sensitive information or allow harmful code to run on a server. The report says researchers found the problem before any known attacker used the exploit.

The background

PHP was created in 1994. It still supports many older websites. The problem began as an out-of-memory error. In simple terms, the program did not have enough space for some data. Such errors often make software crash. They do not always look like an immediate security danger. The bug had been reported to the PHP Foundation in 2024, but it remained unresolved.

Where AI entered

Calif, a cybersecurity company, audited PHP with support from OpenAI, an AI company. Within two weeks, a researcher used AI tools to reason through the old problem. The work showed that the crash could become a much more serious weakness. The researchers did not rely on AI alone. They already understood PHP and computer security. They checked the result and reported it to the PHP Foundation, the group supporting PHP.

Why this matters

The story is about technical debt. Technical debt means old maintenance work that keeps piling up. A system can work for years while its hidden problems grow. AI may help experts find those problems faster. That is useful for defenders. It also creates pressure, because attackers and auditors may examine old software more quickly.

Finding a weakness is not the same as fixing it. A repair must close every important path to the problem. It must also keep the software working. The report describes this as a basic imbalance between discovery and repair. Security teams often need more time for the second task.

What is confirmed

The PHP Foundation received the warning. The report says four engineers spent two months preparing a patch. As far as the report could establish, no attacker had used PHP2Shell before the patch was released.

The story also drew 160 points and 66 comments on Hacker News, a technology discussion site. Those numbers measure community attention. They do not prove that the original report is correct.

The article points to formal verification as one possible defense. In that approach, an AI system might propose a patch and a proof. A separate checker would test the proof. The checker matters because language models can make mistakes.

What remains unknown and what comes next

It is not clear how many other old software bugs can be found this way. It is also unclear how often AI can suggest a safe repair. A successful investigation does not mean that AI can audit every old system automatically.

The next signs to watch are updated PHP releases, similar reviews of other widely used software, and funding for open-source maintenance. Independent checks of AI-generated patches will matter too. PHP2Shell shows that AI can speed up discovery. It does not remove the human work of deciding, repairing, and verifying.

💬 What HN commenters see when AI exposes old software debt

The discussion balances optimism that AI will uncover long-hidden weaknesses against concern that AI-assisted fixes may add even more complexity. These are commenters’ arguments and expectations, not established measurements.

  • Some commenters expect AI-assisted analysis to incidentally uncover hidden backdoors and zero-days. They expect more patches and CVE work in the short term, with a possible long-term benefit of more robust software; that outcome is not yet demonstrated.
  • The KISS idea—keeping features, code, and dependencies no more complicated than necessary—can make systems easier to reason about and reduce their attack surface. But simplicity alone does not guarantee that a system meets its requirements, and KISS has no easy objective metric.
  • Critics say coding LLMs often add layers of validation, error handling, and protection for unlikely cases, choosing a local bandage instead of a cleaner redesign. Missing context, deadlines, and unclear ownership can all encourage bloated code.
  • The counterargument is that defensive checks are not automatically wasteful. New call paths can appear and old assumptions can stop being true, so lower-level validation and explicit, typed, or tested assumptions can prevent vulnerabilities.
  • LLMs can be useful design partners for comparing alternatives, exploring edge cases, and finding holes. Humans still need to choose between patching and refactoring, weigh regression-testing and team constraints, and treat AI effort estimates cautiously.
  • Memory-safe languages can remove a class of memory-corruption bugs, but they do not eliminate logic errors, malicious inputs, or compiler and runtime defects. Rewriting old software or forcing adoption also brings cost, skills, compatibility, and possible new-bug risks.
  • Formal verification, types, and tests can increase confidence, but they cover only the properties and assumptions that were actually defined. Wrong specifications, configurations, or interactions between components can still fail, so large systems still require human understanding and review.

initial digest at 66 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.

🔥 Trending on HN

AI Helped Find a Serious Problem in Old Website Software

📰 Full story: AI Found a Dangerous Crack in Old Software. People Still Had to Fix It.

Researchers used AI to uncover a dangerous problem in old PHP software.

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

Words
PHP

A programming language used to build websites.

technical debt

Old fixing work that has not been finished.

Hacker News

A website where people discuss technology news.

💡 The gist

  • AI helped researchers uncover a serious problem in old PHP software.
  • People reported the problem and released a fix.
  • The case shows why old software needs regular care.

IEEE Spectrum, a technology magazine, reported the story. PHP is a programming language used to build websites. The problem was called PHP2Shell.

At first, the problem looked like a normal crash. A crash makes a program stop. It does not always look dangerous. The bug had been reported in 2024. It stayed unfixed for a long time.

Calif, a cybersecurity company, audited PHP. OpenAI, an AI company, supported the audit. Within two weeks, a researcher used AI tools to study the problem. The work showed a bigger security risk.

AI did not solve the problem alone. The researchers already understood PHP and computer security. They checked the result themselves. Then they reported it to the PHP Foundation, the group supporting PHP.

A patch was made after the report. A patch is an update that fixes software. The article says four engineers worked for two months. It also says no known attacker used the problem before the patch.

This story explains technical debt. Technical debt is old work that still needs fixing. It can grow when software keeps running without enough maintenance. AI can make old problems easier to find. People still need to review and repair them.

Fixing is harder than finding. A fix must close the problem. It must also keep the software working. Tools that check patches carefully may help.

The story received 160 points and 66 comments on Hacker News, a technology discussion site. These numbers show attention. They do not prove the story is true.

Researchers will need to study other old software. They will also need to test whether AI can suggest safe fixes. The main lesson is simple: faster discovery makes regular maintenance more important.

💬 Why fixing old software with AI can help—and hurt

The comments see AI as a way to find old holes, but also warn that careless AI-assisted fixes can make software larger and harder to understand.

  • AI may find hidden backdoors and vulnerabilities that people missed. That could mean more repair work now and stronger software later, but this is a prediction, not a proven result.
  • Using fewer parts, less code, and fewer dependencies can make software easier to understand and reduce places attackers can reach. Still, a simple system may not do everything it needs to do.
  • Coding AI may add too many checks and protections for rare situations, making code bloated. Some of those checks can also protect against future code paths, so people must decide which ones are truly needed.
  • AI can help compare designs and find edge cases. Humans must decide whether to patch or redesign, and must account for testing time and team constraints.
  • Memory-safe languages, types, tests, and proofs help, but they do not remove logic mistakes, harmful inputs, or wrong assumptions. People still need to understand and review the result.

initial digest at 66 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.

🔥 Trending on HN

AI Found a Tiny Gap in an Old Website Tool

📰 Full story: AI Found a Dangerous Crack in Old Software. People Still Had to Fix It.

AI found a hidden gap. People worked together to close it.

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

Words
PHP

A tool people use to make websites.

patch

A software update that fixes a problem.

Hacker News

A website where people talk about technology news.

AI is a computer that helps people think. It found a hidden gap in old website software. The software is called PHP (a tool for making websites). PHP began in 1994. Many websites still use it. The gap was called PHP2Shell. At first, it looked like a program might simply stop. More checking showed a bigger worry. A bad person might use the gap. Calif (a computer-safety company) studied PHP. OpenAI (an AI company) supported the study. Researchers used AI to help with hard checking. They knew PHP and computer safety too. AI did not do the whole job alone. They told the PHP Foundation, PHP’s supporting group. A patch (an update that fixes software) was made. Four engineers worked on it for two months. The report found no known early attack. Hacker News (a technology news site) discussed the story. The record shows 160 points and 66 comments. Those numbers show attention, not truth. Old software needs care, like an old toy. People must keep checking its hidden gaps. IEEE Spectrum reported the story.

💬 AI can find old software holes, but people must guide it

AI may find hidden problems in old programs. But adding too many fixes can make the programs harder to understand.

  • AI might find holes that were hidden for a long time. There may be more fixing to do now, and we do not yet know how much better the software will become.
  • Fewer parts usually make a program easier to check and give attackers fewer doors. But a program that is too simple may not do its job.
  • AI sometimes adds many extra rules. Some rules can protect against future changes, so a person must choose what to keep and whether to fix or rebuild.
  • Safe languages, tests, and proofs help, but they are not magic. AI can be a helper; people must check the design.

initial digest at 66 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.

Sources