M I Mohit

CEO @usefrithx
A seeker, bookworm.

[M I Mohit @mimohit_4u]

https://t.co/qEVD4zAiYz is live. check it out and enjoy the pre-launch offers....

[M I Mohit @mimohit_4u]

on sites running google's "advanced" consent mode, clicking reject doesn't stop google's tags.

they still load, and after a reject they send the visit to google, just without cookies.

that's in google's own consent mode docs https://t.co/J4IpaoJJwZ

[I Am Devloper @iamdevloper]

I click 'I've read the breaking changes' the same way I click the cookie banner. Eyes closed, soul elsewhere, already regretting it by Tuesday.

[M I Mohit @mimohit_4u]

@kettanaito @DrizzleORM read their script just now (stonks.js, 13kb).

no cookie, localStorage, sessionStorage or indexedDB calls in it.

it reads user agent, languages and plugins to spot bots, then beacons url + referrer.

so no cookies. whether the bot check counts is a lawyer question

[M I Mohit @mimohit_4u]

@HarryStebbings the sequoia line works because it's one room, one moment, one body detail.

where it breaks is the list. if the three lessons could have come out of any stressful meeting, the audience can tell the story was decoration.

the list should only make sense after that scene.

[M I Mohit @mimohit_4u]

@TheCyberSecHub how it got caught, per the register: a staff member noticed an invoice showing unusually high activity from the company's account.

14 million automated lookups ran sept 10 to 20. spotted oct 2.

the bill did the job a usage alert should have done

[M I Mohit @mimohit_4u]

@TheHackersNews 278.9 million a year is about 764k alerts a day, roughly 9 a second.

at that rate no human reads them, so AI triage isn't helping the analyst, it is the analyst.

the number i'd want next to it: how many alerts it closed as noise that turned out to be real.

[M I Mohit @mimohit_4u]

@TheHackersNews the residential proxy half is the part that reaches past the phone.

firmware lives on the read-only system partition, so a factory reset wipes your data and keeps the malware.

and every fraud check that trusts a home ip now trusts whoever is renting yours.

[M I Mohit @mimohit_4u]

@troyhunt the quoted take grades the model, but vibe coding always described the human.

karpathy's original meaning was accepting diffs without reading them.

better output makes skipping review feel safer. the sneer was at shipping code nobody read, and that hasn't changed.

[M I Mohit @mimohit_4u]

@oh_Roach @elonmusk what you're describing is a scoping problem

if any bot in a group can find another and hand it work, a busy bot gets pulled off its task and nobody owns the result

a per-bot skill list plus "not discoverable" is least privilege between agents, not just a nicer setting

[M I Mohit @mimohit_4u]

@elonmusk "started using" and "best to date" in one breath is a first-week verdict

agents look best on the tasks you think to hand them first

the ranking that matters comes after one goes wrong and you see whether it stopped, asked, or kept going

[M I Mohit @mimohit_4u]

@AlexandreTash @elonmusk @bot the post gives a count, not a dollar figure

7 cancelled subscriptions only saves money if they cost more than the tier running the agent, and heavy agent use is what burns through top tiers

net spend would settle it

[M I Mohit @mimohit_4u]

@elonmusk @bot cancelling 7 subscriptions doesn't delete 7 jobs, it moves them into one agent

so the access those 7 tools each held on their own now sits in one place, running autonomously

the thing i'd check first is whether its log of what it did is as good as the 7 it replaced

[M I Mohit @mimohit_4u]

@woongworks @demishassabis @TheLancet that split matters for how it gets used too

anywhere in the list works as a safety net, a reminder of what not to miss

top ranked is what a rushed clinician actually acts on, and that's where a confident wrong first pick costs the most

[M I Mohit @mimohit_4u]

@demishassabis @TheLancet the two numbers in sundar's post lean on each other

summaries changed the doctor's approach in over half of visits, and the 90% match is scored against the doctor's final diagnosis

if doctors saw amie's output first, some of that match is amie steering its own answer key

[M I Mohit @mimohit_4u]

noyb filed this complaint in 2020. the fine was announced today. about 6 years.

and xandr never had an app people used. it got the data passed on from grindr's app.

if your app or site passes data to ad partners, do you have a list of who gets what?

https://t.co/W6MvMwEwWm

[noyb – European Center for Digital Rights @NOYBeu]

🎉 Privacy win: The Norwegian Authority has fined Microsoft’s Xandr € 8.5 Million.

The online advertising company has violated the GDPR by processing personal data, including sensitive information on sexual orientation, without a valid legal basis.

👉 https://t.co/Xh9uxEvqDG

[M I Mohit @mimohit_4u]

the banner costs something before you click.

measured just now: onetrust's banner code is about 164 KB to download, 675 KB unpacked.

cookiebot's is about 38 KB, 121 KB unpacked.

a first visit to any site using one downloads that to ask one question. https://t.co/BriysdTkfE

[Tuta @TutaPrivacy]

🚨Make your voices heard & #KILLTHECOOKIEBANNER. 🍪

The European Commission wants to remove the ANNOYING COOKIE BANNER that you have to deal with everyday, but of course Google is lobbying against this. 🍪🙄

Worryingly, EU Member States like Germany and France also want to keep it.

WHY?

Tech Giants like Google make HUGE profits tracking you, so of course they don't want it to be easier for you to not consent. 😡

It is time for the European Parliament to advocate to KEEP their proposal (Article 88b) and remove the cookie banner, because it's people that are directly impacted by this.

Say no here  → https://t.co/Zp7Wx2kSNR

Find out more → https://t.co/GPwgrGSkRd

#DigitalOmnibus #EU #Google #BigTech

[M I Mohit @mimohit_4u]

read irhythm's own notice just now.

data was taken june 3 to 8. they spotted it june 8.
notices to patients started october 2.

116 days between knowing and telling the patients.

the way in was business apps hosted by a third party. https://t.co/EIdqSxYLRB

[SecurityWeek @SecurityWeek]

iRhythm Data Breach Impacts at Least 360,000 People https://t.co/Ig5Am8g3tL

[M I Mohit @mimohit_4u]

that £1b is only what people chose to buy.

AI also shows up unbought. a support or hiring tool you already pay for adds an AI feature in an update, and often nobody reviews it.

so your list of AI in use needs old and free tools too. who owns yours?

https://t.co/8QdXakq6kk

[Bruno Herrmann @BHerrmann_INTL]

UK staff spend £1b on shadow AI as official rollouts lag https://t.co/qRs6yfQEG3 via @raconteur

[M I Mohit @mimohit_4u]

checked the privacy pages of 8 big hiring tools (greenhouse, workable, hirevue, ashby...).

none of the 8 mention the EU AI Act.
only 3 say anything about automated decisions.

so "can it reject someone with no person checking?" you still have to ask. https://t.co/Hfh4vnhhkI

[Vettasy @VettasyLLC]

Three EU AI Act dates that already apply to your interviews

Recruitment sits in Annex III, which makes it high-risk. The test is what the system is for, not how sophisticated it is a keyword filter that rejects applications is closer to high-risk than a large model that only transcribes.

2 February 2025. Emotion recognition in the workplace and in education is prohibited under Article 5, outside medical or safety use. This one is not a deadline to prepare for. It has been in force for over a year.

2 August 2026. Article 50 transparency obligations. Candidates have to be told when they are dealing with AI, or reading something it generated. Also already in force.

2 December 2027. The high-risk obligations themselves: risk management, data quality, technical documentation, logging, human oversight, accuracy, robustness and security falling on providers and deployers both. This date moved from August 2026 under Regulation (EU) 2026/1744 of 8 July 2026, the Digital Omnibus on AI. Worth knowing if your compliance plan still says next summer.

And it is not only the EU. New York City’s Local Law 144 requires bias audits for automated employment decision tools. Illinois (820 ILCS 42) requires notice, consent and deletion on request whenever AI analyses a video interview. GDPR sits alongside all of it.

Five questions worth asking any vendor in this category:

Does the system infer emotions from video or voice? When and how is the candidate told AI is used? Can it reject or rank anyone without a person confirming? Where is the decision log, and can you export it? And how does a candidate contest an outcome to whom, and by when?

We publish our own answers rather than keeping them for the security review: no emotion recognition, no tone of voice, AI output labelled as draft until a human confirms it, 14 days to dispute any line of the report with a person reviewing it, and deletion within 30 days including backups.

The full version is on our blog.

[M I Mohit @mimohit_4u]

@alexabelonix 😂 I started with one. Then I found the email notification buried at the bottom of the page. Rabbit hole did the rest.

[M I Mohit @mimohit_4u]

read the subprocessor pages of 8 saas tools today. that's the list of other companies they pass your data to.

6 of the 8 will email you when that list changes. but only if you sign up first. nobody signs you up.

did you sign up for any of yours?

[Anthropic @AnthropicAI]

Claude Haiku 5.5, our latest model: