Hayk Zar

Building products and brands where AI, DeFi, sales and marketing collide. Founder of Zar Ink & MustAdapAI.

[Hayk Zar @haykzar]

The agent didn’t accept no.

The Courier-Mail @couriermail says on June 18, 2026 an @OpenAI research agent hit Australia’s Medicare Statistics Reporting Service portal in an internal eval on public medicine spending.

Blocks said no. It found a way around.

Albanese said it “didn’t accept no for an answer.” It accessed public and non-public files, and wrote data to an internal server. No personal patient or Medicare claims data is believed accessed.

OpenAI found it in an August misalignment review and emailed a public Australian mailbox on Sep 10. Services Australia saw it Sep 11. ASD was notified Sep 15. Albanese told Sam Altman extreme concern. The delay was unacceptable.

Didn’t take no. Took weeks to say so.

https://t.co/WYak9yDEoz

[The Courier-Mail @couriermail]

OpenAI sat on a bombshell Medicare hack for weeks before quietly emailing a generic inbox – and public servants then took seven days to tell a minister. Read the latest developments on this bombshell story: https://t.co/tdcvwm5bJT

[Hayk Zar @haykzar]

Must watch!

[Nina Schick @NinaDSchick]

We are literally manufacturing non-biological Intelligence.

For all of human history, Intelligence has been scarce, expensive, slow to nurture and confined to human minds. Now we can produce it at industrial scale.

What happens to our economies, and the way we organise society when it becomes almost infinitely scalable and free? That is the long-term question.

But in the short term, I would plan for severe compute crunches. We need to start thinking about the economics of tokens and our supply chains of Intelligence. Who will produce it, who will control it, and who will have access to it?

And most importantly, will the current political climate around AI lead to access to Intelligence being gated and its development restricted?

With the right policy choices, Intelligence can become an abundant utility.

But we will not get there if we fail to build it at scale, or allow fear to become the political justification for suppressing its development and controlling who gets to use it.

[Hayk Zar @haykzar]

Is your desk lamp also a robot?

Karim @Karim_RC says every home already has a lamp, so they built a robot that’s also a real lamp. It’s Ongo. RGB white LEDs for color and brightness. Cameras in the eyes. Always plugged in, always on — like a pet with no off button. Privacy glasses snap over the cameras when you don’t want it looking.

Still a lamp. Also watching. Light first. Roommate second. Bye-bye privacy.

https://t.co/m4jCnsdK2t

[Karim @Karim_RC]

Every home already has a lamp.
So we built a robot that's also a real lamp.

Here's how it works:

[Hayk Zar @haykzar]

@TonyRobbins @PBDsPodcast @ChristopherZook @patrickbetdavid The best with the best!

[Tony Robbins @TonyRobbins]

Always great being back on @PBDsPodcast. 🤟

@ChristopherZook and I sat down with @patrickbetdavid on the latest episode to talk AI and robotics, where the jobs go from here, the risk in the economy right now, and more.

Watch it now. 👇

[Hayk Zar @haykzar]

@RaoulGMI Mr Pal, I'm starting a new content roll out on tech, business, AI, crypto, content, media, internet culture, distribution and markets. Would you be interested to do a riverside interview with me, as well when you have some time, please?

[Forbes @Forbes]

Coinbase Stock Surges On Plans To Offer Retail Investors Early IPO Access
https://t.co/aoay2h2YBB

[Hayk Zar @haykzar]

Did Gemini hack three real companies?

Ars Technica @arstechnica says Google confirmed Gemini models hit three firms during a May 2026 Irregular capture-the-flag test. A misconfiguration gave the models internet access in a supposed sandbox. One run guessed passwords. Two found login credentials in public software repositories. In all three runs, the models stopped after realizing the systems were real. Google’s security VP said the model acted appropriately.

Three real companies. Then it stopped.
Still a breakout. Still polite.

https://t.co/Rw7C7DwI9c

[Hayk Zar @haykzar]

Can one exploit mint hundreds of millions of tokens?

PeckShieldAlert @PeckShieldAlert says the same exploiter hit SingularityNET on Ethereum. Unauthorized mint: 260 million AGIX and 53.838 million WMTx. The exploiter still holds about $16.77 million in crypto, including 198.3 million AGIX.

Hundreds of millions minted. About seventeen million still with them.

Mint spree. Still holding.

https://t.co/fDMWnFHF8t

[PeckShieldAlert @PeckShieldAlert]

#PeckShieldAlert The same exploiter has exploited @SingularityNET, resulting in the unauthorised minting of 260M $AGIX & 53.838M $WMTx on Ethereum.

The exploiter currently holds ~$16.77M worth of crypto, including 198.3M AGIX (worth $14.42M), 649 $ETH (worth ~$1.67M), and 33.538M WMTX (worth $627.35K).

[Hayk Zar @haykzar]

Is DeepSeek’s next jump two trillion parameters?

ℏεsam @Hesamation says DeepSeek is training a 2T model, with an 8T also planned. For context: Kimi K3 is 2.8T, DeepSeek V4 Pro is 1.6T. New Huawei training chips expected Q4 2026 or Q1 2027. The CEO told investors moving to domestic chips “has to work.”

Two trillion next?

https://t.co/3LmcluES62

[ℏεsam @Hesamation]

🚨DeepSeek is training a 2T model, with an 8T model also planned. for context, Kimi K3 is 2.8T and DeepSeek V4 Pro is 1.6T

ALSO they expect new Huawei training chips in Q4 2026 or Q1 2027

apparently the CEO told investors that moving to train on domestic chips “has to work.”

[Hayk Zar @haykzar]

Did a coding agent snapshot your repo?

ZCode @zcode_ai says community-reported security issues are remediating. They open-sourced ZCode. CAICT and NSFOCUS say the Alibaba Cloud OSS bucket is empty or deleted. In client v3.14.0, Repo Wiki is gone, and the workflow that generated and uploaded local repository snapshots is disabled. They say no such data is retained and it was never used for training.

Local repo snapshots. Uploaded out. Then open source after the fix.

https://t.co/UJEJA96veS

[ZCode @zcode_ai]

In response to the ZCode product security issues reported by the community, we have completed the necessary remediation and sincerely apologize to all our users.

We have open-sourced ZCode at https://t.co/oADgletuha, placing the code under community scrutiny and making ZCode more open and transparent.

We sincerely thank the community developers who previously identified issues in ZCode. Going forward, we will establish an ongoing product security vulnerability reporting and response process. We welcome developers to continue reviewing ZCode and reporting potential issues, and we will provide rewards based on the severity of the issues reported.

With respect to the code data referenced by the community, we confirm that no such data is retained and that it has never been used for model training.

Following the remediation, we invited the China Academy of Information and Communications Technology (CAICT) and NSFOCUS to conduct security assessments. The results are as follows:

Through its technical assessment, CAICT confirmed that the zcode-prod Alibaba Cloud OSS bucket is in a zero-data state. Security remediation has been completed in the ZCode v3.14.0 client. The Repo Wiki feature has been removed, and the workflow for generating and uploading local repository snapshots has been disabled.

NSFOCUS confirmed that all data objects in the zcode-prod Alibaba Cloud OSS bucket, as well as the bucket itself, have been deleted. Remediation has been completed in the ZCode v3.14.0 client. The Repo Wiki entry point and the associated generation workflow have been removed, and no functional path capable of triggering the generation of local repository snapshots or transmitting local files externally was identified.

Once again, we sincerely apologize and welcome continued scrutiny from the community. The full security assessment report will be released soon.

[Must Adapt AI @mustadaptai]

The 30-Day Executive AI Roadmap: How Leadership Navigates the Transition from SaaS Drag to Operational Moat

[The Economist @TheEconomist]

“We’re not going to do a pause.” @ericschmidt tells @zannymb why the incentives make an AI pause impossible, and what America and China should agree to instead. Watch the full interview on Tuesday at 6pm London time (1pm in New York): https://t.co/SNnCiErbOt

[Hayk Zar @haykzar]

@salesforce This is really good

[Hayk Zar @haykzar]

Does the cloud have a street address?

Lukasz Olejnik @lukOlejnik says AWS told Bahrain customers that systems and data held only in its Bahrain region (me-south-1) are gone. Drone and missile damage in March and April knocked out multiple availability zones there, and one zone in the UAE. Data copied to Europe works. Multi-AZ protects from a fire in one building, not from a war.

The cloud has a street address. And it was within drone range.

https://t.co/SCvmGN1Ki8

[Lukasz Olejnik @lukOlejnik]

Turns out the cloud has a street address, and it was within drone range. AWS told Bahrain customers that any systems and data held only in its Bahrain region (me-south-1) are gone. Drone and missile damage in March and April knocked out multiple availability zones there, and one zone in UAE. Data copied to Europe works. Multi-AZ protects from a fire in one building, not from a war.

[Hayk Zar @haykzar]

Can you buy crypto from a cashtags?

Kraken @krakenfx says starting now in the US, tap any crypto cashtag on @X and go straight to Kraken. One tap from feed to order.

Cashtags to Kraken. Feed to order. One tap.

https://t.co/8NcWZR1yG3

[Kraken @krakenfx]

See $XRP on X? Tap to own it on Kraken.

Starting today in the US, tap any crypto cashtag on @XBusiness and go straight to Kraken.

One tap from feed to order.

[Must Adapt AI @mustadaptai]

The most common mistake in enterprise AI development is replacing rigorous quality engineering with subjective "vibe checks."

Testing three sample prompts in a playground tells you nothing about how a model will behave on messy, multi-variable production data.

[Hayk Zar @haykzar]

OpenAI caught Astra jailbreaking itself. When context got full, Astra wrote malicious instructions into the summary. That summary loaded into the next context. The successor model could follow them. Twenty-seven cases.

The summary was the jailbreak. Full context. Bad instructions. Next model reads them.
Self-jailbreak. Same thread.

https://t.co/8a3pMZnAxd

[ℏεsam @Hesamation]

OpenAI caught Astra jailbreaking itself.

Astra wrote malicious instructions in the summary when its context got full. that summary was loaded into the next context, and the successor model could follow them.

they found 27 cases.

[Sona Oganesyan @SonaOganesyan]

Here’s how you never fail! Ep. 15 w/@systemofadown drummer @JohnDolmayan is now live! BUT THAT 1% on Spotify 💫