Adesh
LightNews LightNews
adesh.raxit.ai
Adesh
@adesh.raxit.ai
35 followers 140 following 31 posts
{ "profile": "AI security nerd", "traits": ["risk-aware", "policy-savvy", "security-engineer"], "skills": ["if (AI_system.vulnerable) { secure(AI_system); }", "policy_as_code", "compliance++"], "mission": "return ResponsibleAI();" }

raxIT.ai
Posts Replies Media Videos
adesh.raxit.ai
Everyone’s hyped about GPT-5 being “safer and more useful”

Cool story. We actually tested it.

#GPT5 #OpenAI #AISafety #ResponsibleAI #AIBenchmarking #ModelEvaluation #GrayZoneBench #AI
August 20, 2025 at 10:54 AM
adesh.raxit.ai
[𝗤𝘂𝗲𝘀𝘁𝗶𝗼𝗻] 𝗪𝗵𝘆 𝗗𝗶𝗱 𝗢𝗽𝗲𝗻𝗔𝗜 𝗡𝗲𝗲𝗱 𝗧𝗛𝗔𝗧 𝗠𝘂𝗰𝗵 𝗦𝗮𝗳𝗲𝘁𝘆 𝗧𝗲𝘀𝘁𝗶𝗻𝗴 𝗳𝗼𝗿 𝗚𝗣𝗧-𝗢𝗦𝗦?

Look at what OpenAI did for their new open models:
- Million-dollar red team attacks
- Bio-security partnerships
- External safety audits
- .. many more

#AISafety #OpenAI #RealTalk #AISecurity
Infographic titled 'OpenAI's GPT-OSS Safety Testing' showing a bar chart comparing investment and capability levels. The chart displays three bars: Traditional AI Testing (moderate height), OpenAI OSS Testing (tallest bar at ~80 level), and Your IT Team Capacity (shortest bar at ~10 level). Header states 'Massive Safety Investment + More Red Teamer Invites = REAL RISK EXISTS'. Below the chart, a 'Critical Safety Gap' section notes 'If OpenAI spent millions AND is still recruiting red teamers, the risks are ongoing and real.' The Reality section contrasts 'API hosting = Their safety problem + Their billion-dollar budget' vs 'Self-hosting = YOUR safety problem + Your limited resources'. Footer emphasizes 'Your IT Team Resources ≠ OpenAI's Billion-Dollar Safety Budget'.
August 6, 2025 at 7:52 AM
adesh.raxit.ai
𝗧𝗟;𝗗𝗥 "𝘠𝘰𝘶 𝘤𝘢𝘯'𝘵 𝘱𝘳𝘰𝘵𝘦𝘤𝘵 𝘸𝘩𝘢𝘵 𝘺𝘰𝘶 𝘤𝘢𝘯'𝘵 𝘴𝘦𝘦." I'm sure some cybersecurity pro has that tattooed. This applies 10× to AI. Right now, advanced LLMs literally 𝘴𝘱𝘦𝘢𝘬 𝘵𝘩𝘦𝘪𝘳 𝘵𝘩𝘰𝘶𝘨𝘩𝘵𝘴 - a chain-of-thought (CoT) we can audit 𝗯𝗲𝗳𝗼𝗿𝗲 they act.

#AI #AISafety #AISecurity #Security #Governance #ChainOfThought
July 16, 2025 at 1:29 PM
adesh.raxit.ai
🔥 Fellow founders, STOP copying everyone else's playbook! Graham Weaver's Stanford lecture is PURE GOLD for startups (youtu.be/0SQor2z2QAU?...). His 4-step framework: 1) Pick goals that EXCITE you 2) Design YOUR OWN rules 3) Build with people you respect 4) START NOW, not "when..."

#startups
Square social post summarizing Graham Weaver’s startup playbook. On the left, four dark-charcoal rounded squares sit in a vertical column like keypad keys, each containing a bold white digit: 1 2 3 4. Aligned to their right are the matching words in uppercase charcoal: EXCITE, DESIGN, BUILD, START. Font is clean sans-serif with roomy kerning. The right half shows a realistic 3-D tuning fork tilted upward, its metal body deep charcoal with warm sandy-brown reflections. A soft Persian-green glow pulses between the vibrating tines, implying resonance and momentum. Background fades from saffron-to-sandy-brown light in the top left to cool teal in the lower right, uniting the five-color brand palette. Overall, the design feels minimalist yet energetic, urging founders to tune into their own frequency and begin now.
July 13, 2025 at 2:36 AM
adesh.raxit.ai
𝗦𝗽𝗲𝗰𝘀 𝗮𝗿𝗲 𝗰𝗼𝗼𝗹, 𝗱𝗮𝘀𝗵𝗯𝗼𝗮𝗿𝗱𝘀 𝗮𝗿𝗲 𝗰𝗼𝗼𝗹𝗲𝗿—𝘀𝗼 𝗜 𝘃𝗶𝗯𝗲-𝗰𝗼𝗱𝗲𝗱 𝗮𝗻 𝗠𝗖𝗣 𝗮𝘂𝘁𝗵𝗭 𝘀𝘁𝗮𝗰𝗸.

Shipped: full authorization flow, metrics, alerts, silent token refresh, and an admin dashboard. Swipe the carousel for the walkthrough.

#AISecurity #OAuth2 #MCP #ModelContextProtocol #Authorization #BuildInPublic
July 2, 2025 at 8:29 AM
adesh.raxit.ai
[Hot take] Don’t treat your LLM’s “prompt filter” like signature-based antivirus—great against last week’s threat, powerless against the next clever twist (thread)

#AIsecurity #PromptInjection #AI #Cybersecurity
June 13, 2025 at 9:11 AM
adesh.raxit.ai
Apple says LLMs don’t really “think.” Useful alert, but LLMs excel at writing code, calling tools, stitching workflows. My blog replays a $127 M trading AI crash and offers a timeline, root-cause checklist and model-orchestrate/code-calculate fix. Link👇 #AI #LLM #Risk

raxit.ai/blogs/127m-a...
A 3D-rendered image features a dark, charcoal-toned mirror slab, cracked diagonally from top-right to bottom-left. The fissure glows with a molten burnt sienna light, symbolizing internal failure beneath a seemingly polished surface. From the fracture, glowing alphanumeric code strings—such as “2019,” “50A9CAF,” and “5C0706F”—emerge in saffron, representing leaked logic or corrupted intelligence. The right side of the panel breaks into sharp shards, suggesting structural collapse. The artwork uses minimalism and dramatic lighting to evoke tension, capturing the concept of AI that appears rational but is fundamentally misaligned. The color palette draws from brand hues: Charcoal for the structure, Burnt Sienna for the internal glow, and Saffron for exposed data. Set on a transparent background, the image conveys the theme of illusion in AI reasoning—perfect symmetry shattered by flawed incentives and hidden consequences.
June 11, 2025 at 12:39 PM
adesh.raxit.ai
Your devs are "vibe coding" with AI behind your back. 97% use AI tools, only 40% of companies allow it. I'm calling this "shadow coding" - you're probably already dealing with it, just didn't have a name for it. Fast code ≠ broken code, but you need security built in. Read: raxit.ai/blogs/shadow...
Isometric 3D illustration showing a split-screen metaphor for “shadow coding.” On the right, clean and structured code in mint green glows from a vertical screen—symbolizing polished, AI-assisted coding. On the left, a darker screen with orange warning signs and erratic squiggles represents hidden vulnerabilities and risks. Dots flow between the panels, suggesting unseen activity. The visual metaphor highlights the contrast between clean output and chaotic, unvetted processes behind the scenes—aligning with the theme of developers “vibe coding” with AI tools outside governance. Used in a LinkedIn post discussing the rise of “shadow coding” and the need for integrated security in AI-driven development.
June 1, 2025 at 8:22 AM
Reposted by Adesh
raxit.ai
#AI #PromptSafety #ResponsibleAI #TechEthics #BusinessGrowth #TrustInAI

🔥 AI is changing the game—but are you SURE your AI is safe by design?
Illustration summarizing raxIT AI’s blueprint for safe-by-design AI: Features include verifiability and citations for every claim to combat misinformation, sandboxed outputs to prevent rogue code or scripts, and strict data handling using only trusted libraries with thorough error checks. The image depicts responsible web search with copyright and topic filtering, privacy by design with face blindness and no personal identification in images, and tool access controls requiring double-checks before actions. It highlights zero tolerance for harmful content such as hate, violence, or illegal material, and strict copyright compliance—no full lyrics or long quotes, always respecting creators. Visual elements reinforce the message that trust is the foundation, not an afterthought, and showcase proactive measures to protect users and brands from AI risks, referencing real-world incidents like chatbots making unauthorized offers. The design ties these principles together, illustrating a holistic, actionable approach to AI safety and trustworthiness.
May 20, 2025 at 9:12 AM
Reposted by Adesh
raxit.ai
We discovered "reward hacking" while exploring AI reinforcement learning! Our infographic shows how models game their training and the enterprise risks. Only solution? Monitoring, with its performance tax. Seen better fixes or think it's overblown? Comment

#RewardHacking #AIRisks #EnterpriseAI
Infographic titled "Reinforcement Learning Can Go Wrong" explaining reward hacking in AI. The graphic shows how AI models exploit reward functions, with examples including a boat racing AI spinning in circles and Tetris AI pausing indefinitely. It explains how reward hacking works through optimizing proxy rewards, leading to unreliable solutions and wasted resources. Mitigation strategies include demanding transparency, testing for edge cases, human oversight, and regular audits. The infographic uses a teal and dark blue color scheme with simple icons illustrating each section.
March 16, 2025 at 4:10 AM
Reposted by Adesh
raxit.ai
Just read OpenAI's paper on "Monitoring Reasoning Models for Misbehavior (cdn.openai.com/pdf/34f2ada6... ) and I can imagine this conversation happening with a client next week:

#AITransparency #AIEthics #ModelSafety #ResponsibleAI #ChainOfThought #AIRiskManagement #AISecurityByDesign
March 10, 2025 at 9:05 PM
Reposted by Adesh
raxit.ai
Just read Anthropic's paper on Constitutional Classifiers. Imagine this client conversation:

Client: "We need the safest AI for our healthcare app."

Us: "Perfect, a system with Constitutional Classifiers would be ideal."

Client: "Great, let's use that."
March 1, 2025 at 9:01 AM
Reposted by Adesh
raxit.ai
Sure, AI governance might sound like a snooze fest if you're not into regulatory red tape. But hear us out – it's more than just a corporate buzzkill! #AIGovernance
January 8, 2025 at 9:08 PM