@BarakSeri קצת מוזר לומר את זה על ראש ממשלה שהיה שם מספר פעמים בשנה כנראה יותר מכל מנהיג אי פעם.
כל כך הרבה פעמים שמענו שהיחסים עכורים. ואז שוב הגיע ביצוע מושלם של תיאום.
חפשו ברשת "היחסים בין ישראל לארהב ברמת שפל..."
גוגל לא שוכח
History doesn’t always repeat itself, but sometimes it ships a new version.
This month, Anthropic’s model-access shutdown made something clear: the old crypto wars are back — only this time, the battlefield is AI.
History -
In the 1990s, the U.S. government restricted the export of strong encryption. SSL was becoming the foundation of online commerce, but browsers outside the U.S. were often limited to weaker "export-grade" crypto.
The reason was national security: strong encryption protects privacy, banks, companies, and commerce — but it can also make intelligence and law-enforcement work harder.
One "creative" compromise was the Clipper Chip in 1993: allow strong encryption, but keep backup keys in escrow so the government could access communications with legal authorization.
To some policymakers, it sounded like a balance between security and privacy. To many security experts, it looked like a built-in backdoor.
The idea failed, but the legacy of weakened crypto stayed with us. Years later, attacks like FREAK and Logjam exploited old export-grade crypto support in SSL/TLS.
Fast-forward to AI.
In June 2026, according to Anthropic, the U.S. government issued an export-control directive requiring the company to suspend access to Fable 5 and Mythos 5 for foreign nationals, including foreign-national employees. Anthropic said it disabled access broadly to comply.
Anthropic also said it had already tried a "defense in depth" approach: safeguards, red-teaming, monitoring, and temporary data retention to detect misuse.
That is the interesting analogy.
Maybe the modern version of "key escrow" is not a backup encryption key.
Maybe it is monitored access, trusted users, logging, retention, and controlled deployment.
But the old lesson still matters:
When governments try to control powerful general-purpose technology, the workaround often becomes the real story.
And if the control mechanism is badly designed, its side effects can last longer than the threat it was meant to solve.
Sources: Anthropic, WIRED, White House Archives, NIST, Federal Register / GovInfo, NVD, Logjam researchers.
I gave Claude Code a pair of eyes 👀
It writes firmware, flashes it to a real dev board, watches the LEDs through a camera, and fixes its own bugs — fully autonomous.
Here's the rig and how the loop closes 🧵
Under the hood it's three skills:
• /goal — defines the desired hardware state (the target LED pattern)
• /loop — periodically snaps a camera frame to check the board actually matches that goal
• a custom skill I wrote to make driving the camera easy
Did the Claude team picture this when they shipped /goal and /loop? 😄 @AnthropicAI@claudeai
Here is how the autonomous loop works:
1️⃣ The Goal: I use the /goal skill to define exactly what the firmware should achieve (e.g., a specific LED blinking pattern).
2️⃣ The Action: Claude writes the code and flashes it directly to the board via a J-Link debugger.
3️⃣ The Vision: Using a custom skill I wrote, Claude triggers an overhead inspection camera to snap a photo of the running hardware.
4️⃣ The Loop: It uses the /loop skill to continuously monitor the device, analyzing the real-world visual state against the desired baseline.
5️⃣ The Proof: Once validated, Claude reports the status directly to my Telegram, complete with data graphs and the captured media as proof.
@elgarat_dani הערבית לא נוכחת כמו שאתה מנסה לצייר.
יהודי�� שכלו ממדינות ערב לא גאים בשפה הזו ולא מחצינים אותה.
את אמא שלי מעולם לא שמעתי אומרת אינשאללה וגם לה מפריע שיש שירים בערבית בסופר.
אני לא חושב שזה בסדר לחסום את לאסי בשכונה שלה למטה כמו שאני חושב שזה לא בסדר לחסום כל אחד בחדר מדרגות שלו