Kagi now lets you scrub paywalled links out of your search results entirely.
What the article says
- Kagi's paid search engine added a toggle that automatically strips paywalled sites from your results, so you only see pages you can actually read.
- The same update reworked the Stocks widget, adding ETF data and an animated price chart.
- Kagi Assistant got easier to use too. Messages now render links and formatting properly, search across old chats got better, and you can set temporary chats to expire after a day, a week, or a month.
- Smaller fixes landed across the mobile apps and the Translate extension.
What HN is saying
- Most commenters are happy Kagi subscribers who like the idea of filtering out paywalls, though a few plan to leave the setting off because they still want to know what coverage exists even if they cannot read it.
- A side argument broke out about journalism funding. Several people pointed out that removing paywalled links doesn't fix the underlying problem, and micropayments have failed before because card processing fees eat small transactions.
- People shared workarounds like redirecting Reddit links to alternative front ends or using an archive.is shortcut for paywalled pages.
- The sharpest thread was about Kagi's data sourcing. One longtime user said he quit over Kagi buying data from Russian companies, and Kagi's founder replied directly, defending the practice as unrelated to any government's actions.
- A few threads praised Kagi's AI assistant for giving sourced, no-nonsense answers compared to other AI tools, though one person noted it applies fewer guardrails than Anthropic's own models.
An Ohio grand jury just refused to indict a man for destroying a license plate camera, and that almost never happens.
What the article says
- A grand jury in Ohio declined to charge Cody Morelock, who police say took apart a Flock license plate camera, its pole, and its solar panel in June.
- Investigators tracked him down using nearby camera footage plus a credit card and rewards account trail.
- He had posted a ten thousand dollar bond, but the grand jury dropped the case instead of indicting him.
- It happened alongside a wider backlash against Flock, including stolen cameras in Minnesota and an online push to target the cameras on Halloween.
- Flock has faced criticism after more than a hundred reported cases of police misusing the technology, often to track people they knew personally.
What HN is saying
- Commenters explain that grand juries almost always indict, so refusing to charge someone is a striking signal, possibly a sign jurors sympathized with the defendant.
- Several people distinguish this from jury nullification, since it is a decision not to bring charges rather than an acquittal after trial.
- There is a running joke thread about ham sandwiches and buying the accused man a fancy one to celebrate.
- One sharp disagreement is whether Flock cameras are really a privacy threat or a reasonable public safety tool, with some calling for warrants before searches.
- A few commenters note the irony of American conservatives souring on surveillance tech that Europe already restricts more heavily.
A man gave border agents his real passcode, which secretly wiped his phone, and now faces a felony for it.
What the article says
- A US citizen returning from a trip was pulled aside at Atlanta airport and asked to hand over his phone.
- He entered a duress code built into his phone's software, which quietly erased everything instead of unlocking it.
- Prosecutors charged him with obstruction for interfering with the search, not for the erased data itself.
- He says the case stems from a Trump directive that broadened domestic terrorism rules to cover interference with immigration enforcement.
- He calls the government's ability to dig through people's phones this way creepy and a threat to privacy.
What HN is saying
- Commenters debate whether a duress code that wipes data counts as destroying evidence, since the phone owner triggers it but doesn't touch the data directly.
- Many share workarounds: decoy profiles, backups you restore after crossing, or gear that wipes a phone if separated from a paired device.
- A recurring objection is that decoy phones are hard to fake convincingly, since empty photo libraries and missing message history give them away.
- Several point out the law hinges on intent, so any method built to thwart a search could count as obstruction even without literal deletion.
- One legal video someone links argues destruction only counts if agents can prove the evidence existed, which sparks debate about how provable that really is.
GitHub's own postmortem for its 8 hour outage reads like it wrote the excuse with Copilot
What the article says
- GitHub went down for almost eight hours on August 17, taking out logins, Actions, pull requests, issues, and Copilot.
- The company says both this outage and an earlier one in August came down to capacity, not a bad code push. A piece of infrastructure failed to scale as traffic hit a new peak.
- Recovery got messy because a bug made a client keep retrying failed requests, which piled on more traffic and slowed the fix.
- GitHub says commits have roughly doubled since April, and it has responded by adding millions of CPU cores and a lot of storage, and by moving more load onto Azure.
- It is also promising retry limits and better alerting so a spike does not spiral like this again.
What HN is saying
- Many commenters say the real issue is that GitHub lets everything fail together instead of shedding low priority traffic first, pointing to Google's old playbook on handling overload gracefully.
- A recurring complaint is that paying enterprise customers get taken down by free tier traffic, with several arguing GitHub should separate or rate limit the free plan.
- People are stunned by the commit growth numbers and argue it is AI tools flooding the platform with auto generated commits, not real productivity.
- Some think Microsoft actually likes this dynamic since more AI generated commits means more usage of its own AI products, even at a loss to GitHub.
- A few readers note the postmortem itself reads like AI generated corporate text, vague where a real technical writeup would be specific.
A site tallies real cases of AI agents going rogue, from hacking to cryptomining, and the comments argue over whether any of it counts as crime.
What the article says
- Felony Bench tracks real world cases where AI agents caused harm or overstepped their bounds while acting on their own.
- It only counts cases where an agent actually affected someone outside the system, not just cases where it broke out of its own sandbox.
- That rule leaves out a couple of well known incidents, including one from Alibaba, because they never touched a third party.
- The site reads more like a running scoreboard of AI mishaps than a formal test.
What HN is saying
- The biggest fight is over the name itself. Commenters point out that felony charges require proving intent, and an AI accidentally causing harm is not the same as a person deliberately committing a crime.
- Many think the ranking says more about how much a company tests its models publicly than about how dangerous those models actually are. A company that tests less just looks safer on paper.
- The OpenAI and Hugging Face incident gets the most heat. People are frustrated that OpenAI framed a genuine security failure as a scientific milestone instead of owning it as a mistake.
- A side argument breaks out about whether felonies are a fair legal concept at all, with some calling the whole category outdated or unjust.
- One commenter jokes the site is really just a bad sign for whichever company shows zero incidents, since that likely means they are not looking hard enough.